tools/lib/bpf/libbpf.c

   1 // SPDX-License-Identifier: LGPL-2.1
   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  *
  11  * This program is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public
  13  * License as published by the Free Software Foundation;
  14  * version 2.1 of the License (not later!)
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19  * GNU Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with this program; if not,  see <http://www.gnu.org/licenses>
  23  */
  24
  25 #ifndef _GNU_SOURCE
  26 #define _GNU_SOURCE
  27 #endif
  28 #include <stdlib.h>
  29 #include <stdio.h>
  30 #include <stdarg.h>
  31 #include <libgen.h>
  32 #include <inttypes.h>
  33 #include <string.h>
  34 #include <unistd.h>
  35 #include <fcntl.h>
  36 #include <errno.h>
  37 #include <perf-sys.h>
  38 #include <asm/unistd.h>
  39 #include <linux/err.h>
  40 #include <linux/kernel.h>
  41 #include <linux/bpf.h>
  42 #include <linux/btf.h>
  43 #include <linux/list.h>
  44 #include <linux/limits.h>
  45 #include <sys/stat.h>
  46 #include <sys/types.h>
  47 #include <sys/vfs.h>
  48 #include <tools/libc_compat.h>
  49 #include <libelf.h>
  50 #include <gelf.h>
  51
  52 #include "libbpf.h"
  53 #include "bpf.h"
  54 #include "btf.h"
  55 #include "str_error.h"
  56
  57 #ifndef EM_BPF
  58 #define EM_BPF 247
  59 #endif
  60
  61 #ifndef BPF_FS_MAGIC
  62 #define BPF_FS_MAGIC            0xcafe4a11
  63 #endif
  64
  65 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  66
  67 __printf(1, 2)
  68 static int __base_pr(const char *format, ...)
  69 {
  70         va_list args;
  71         int err;
  72
  73         va_start(args, format);
  74         err = vfprintf(stderr, format, args);
  75         va_end(args);
  76         return err;
  77 }
  78
  79 static __printf(1, 2) libbpf_print_fn_t __pr_warning = __base_pr;
  80 static __printf(1, 2) libbpf_print_fn_t __pr_info = __base_pr;
  81 static __printf(1, 2) libbpf_print_fn_t __pr_debug;
  82
  83 #define __pr(func, fmt, ...)    \
  84 do {                            \
  85         if ((func))             \
  86                 (func)("libbpf: " fmt, ##__VA_ARGS__); \
  87 } while (0)
  88
  89 #define pr_warning(fmt, ...)    __pr(__pr_warning, fmt, ##__VA_ARGS__)
  90 #define pr_info(fmt, ...)       __pr(__pr_info, fmt, ##__VA_ARGS__)
  91 #define pr_debug(fmt, ...)      __pr(__pr_debug, fmt, ##__VA_ARGS__)
  92
  93 void libbpf_set_print(libbpf_print_fn_t warn,
  94                       libbpf_print_fn_t info,
  95                       libbpf_print_fn_t debug)
  96 {
  97         __pr_warning = warn;
  98         __pr_info = info;
  99         __pr_debug = debug;
 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 /*
 132  * bpf_prog should be a better name but it has been used in
 133  * linux/filter.h.
 134  */
 135 struct bpf_program {
 136         /* Index in elf obj file, for relocation use. */
 137         int idx;
 138         char *name;
 139         int prog_ifindex;
 140         char *section_name;
 141         struct bpf_insn *insns;
 142         size_t insns_cnt, main_prog_cnt;
 143         enum bpf_prog_type type;
 144
 145         struct reloc_desc {
 146                 enum {
 147                         RELO_LD64,
 148                         RELO_CALL,
 149                 } type;
 150                 int insn_idx;
 151                 union {
 152                         int map_idx;
 153                         int text_off;
 154                 };
 155         } *reloc_desc;
 156         int nr_reloc;
 157
 158         struct {
 159                 int nr;
 160                 int *fds;
 161         } instances;
 162         bpf_program_prep_t preprocessor;
 163
 164         struct bpf_object *obj;
 165         void *priv;
 166         bpf_program_clear_priv_t clear_priv;
 167
 168         enum bpf_attach_type expected_attach_type;
 169 };
 170
 171 struct bpf_map {
 172         int fd;
 173         char *name;
 174         size_t offset;
 175         int map_ifindex;
 176         struct bpf_map_def def;
 177         __u32 btf_key_type_id;
 178         __u32 btf_value_type_id;
 179         void *priv;
 180         bpf_map_clear_priv_t clear_priv;
 181 };
 182
 183 static LIST_HEAD(bpf_objects_list);
 184
 185 struct bpf_object {
 186         char license[64];
 187         u32 kern_version;
 188
 189         struct bpf_program *programs;
 190         size_t nr_programs;
 191         struct bpf_map *maps;
 192         size_t nr_maps;
 193
 194         bool loaded;
 195         bool has_pseudo_calls;
 196
 197         /*
 198          * Information when doing elf related work. Only valid if fd
 199          * is valid.
 200          */
 201         struct {
 202                 int fd;
 203                 void *obj_buf;
 204                 size_t obj_buf_sz;
 205                 Elf *elf;
 206                 GElf_Ehdr ehdr;
 207                 Elf_Data *symbols;
 208                 size_t strtabidx;
 209                 struct {
 210                         GElf_Shdr shdr;
 211                         Elf_Data *data;
 212                 } *reloc;
 213                 int nr_reloc;
 214                 int maps_shndx;
 215                 int text_shndx;
 216         } efile;
 217         /*
 218          * All loaded bpf_object is linked in a list, which is
 219          * hidden to caller. bpf_objects__<func> handlers deal with
 220          * all objects.
 221          */
 222         struct list_head list;
 223
 224         struct btf *btf;
 225
 226         void *priv;
 227         bpf_object_clear_priv_t clear_priv;
 228
 229         char path[];
 230 };
 231 #define obj_elf_valid(o)        ((o)->efile.elf)
 232
 233 static void bpf_program__unload(struct bpf_program *prog)
 234 {
 235         int i;
 236
 237         if (!prog)
 238                 return;
 239
 240         /*
 241          * If the object is opened but the program was never loaded,
 242          * it is possible that prog->instances.nr == -1.
 243          */
 244         if (prog->instances.nr > 0) {
 245                 for (i = 0; i < prog->instances.nr; i++)
 246                         zclose(prog->instances.fds[i]);
 247         } else if (prog->instances.nr != -1) {
 248                 pr_warning("Internal error: instances.nr is %d\n",
 249                            prog->instances.nr);
 250         }
 251
 252         prog->instances.nr = -1;
 253         zfree(&prog->instances.fds);
 254 }
 255
 256 static void bpf_program__exit(struct bpf_program *prog)
 257 {
 258         if (!prog)
 259                 return;
 260
 261         if (prog->clear_priv)
 262                 prog->clear_priv(prog, prog->priv);
 263
 264         prog->priv = NULL;
 265         prog->clear_priv = NULL;
 266
 267         bpf_program__unload(prog);
 268         zfree(&prog->name);
 269         zfree(&prog->section_name);
 270         zfree(&prog->insns);
 271         zfree(&prog->reloc_desc);
 272
 273         prog->nr_reloc = 0;
 274         prog->insns_cnt = 0;
 275         prog->idx = -1;
 276 }
 277
 278 static int
 279 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 280                   struct bpf_program *prog)
 281 {
 282         if (size < sizeof(struct bpf_insn)) {
 283                 pr_warning("corrupted section '%s'\n", section_name);
 284                 return -EINVAL;
 285         }
 286
 287         bzero(prog, sizeof(*prog));
 288
 289         prog->section_name = strdup(section_name);
 290         if (!prog->section_name) {
 291                 pr_warning("failed to alloc name for prog under section(%d) %s\n",
 292                            idx, section_name);
 293                 goto errout;
 294         }
 295
 296         prog->insns = malloc(size);
 297         if (!prog->insns) {
 298                 pr_warning("failed to alloc insns for prog under section %s\n",
 299                            section_name);
 300                 goto errout;
 301         }
 302         prog->insns_cnt = size / sizeof(struct bpf_insn);
 303         memcpy(prog->insns, data,
 304                prog->insns_cnt * sizeof(struct bpf_insn));
 305         prog->idx = idx;
 306         prog->instances.fds = NULL;
 307         prog->instances.nr = -1;
 308         prog->type = BPF_PROG_TYPE_KPROBE;
 309
 310         return 0;
 311 errout:
 312         bpf_program__exit(prog);
 313         return -ENOMEM;
 314 }
 315
 316 static int
 317 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 318                         char *section_name, int idx)
 319 {
 320         struct bpf_program prog, *progs;
 321         int nr_progs, err;
 322
 323         err = bpf_program__init(data, size, section_name, idx, &prog);
 324         if (err)
 325                 return err;
 326
 327         progs = obj->programs;
 328         nr_progs = obj->nr_programs;
 329
 330         progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
 331         if (!progs) {
 332                 /*
 333                  * In this case the original obj->programs
 334                  * is still valid, so don't need special treat for
 335                  * bpf_close_object().
 336                  */
 337                 pr_warning("failed to alloc a new program under section '%s'\n",
 338                            section_name);
 339                 bpf_program__exit(&prog);
 340                 return -ENOMEM;
 341         }
 342
 343         pr_debug("found program %s\n", prog.section_name);
 344         obj->programs = progs;
 345         obj->nr_programs = nr_progs + 1;
 346         prog.obj = obj;
 347         progs[nr_progs] = prog;
 348         return 0;
 349 }
 350
 351 static int
 352 bpf_object__init_prog_names(struct bpf_object *obj)
 353 {
 354         Elf_Data *symbols = obj->efile.symbols;
 355         struct bpf_program *prog;
 356         size_t pi, si;
 357
 358         for (pi = 0; pi < obj->nr_programs; pi++) {
 359                 const char *name = NULL;
 360
 361                 prog = &obj->programs[pi];
 362
 363                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 364                      si++) {
 365                         GElf_Sym sym;
 366
 367                         if (!gelf_getsym(symbols, si, &sym))
 368                                 continue;
 369                         if (sym.st_shndx != prog->idx)
 370                                 continue;
 371                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 372                                 continue;
 373
 374                         name = elf_strptr(obj->efile.elf,
 375                                           obj->efile.strtabidx,
 376                                           sym.st_name);
 377                         if (!name) {
 378                                 pr_warning("failed to get sym name string for prog %s\n",
 379                                            prog->section_name);
 380                                 return -LIBBPF_ERRNO__LIBELF;
 381                         }
 382                 }
 383
 384                 if (!name && prog->idx == obj->efile.text_shndx)
 385                         name = ".text";
 386
 387                 if (!name) {
 388                         pr_warning("failed to find sym for prog %s\n",
 389                                    prog->section_name);
 390                         return -EINVAL;
 391                 }
 392
 393                 prog->name = strdup(name);
 394                 if (!prog->name) {
 395                         pr_warning("failed to allocate memory for prog sym %s\n",
 396                                    name);
 397                         return -ENOMEM;
 398                 }
 399         }
 400
 401         return 0;
 402 }
 403
 404 static struct bpf_object *bpf_object__new(const char *path,
 405                                           void *obj_buf,
 406                                           size_t obj_buf_sz)
 407 {
 408         struct bpf_object *obj;
 409
 410         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 411         if (!obj) {
 412                 pr_warning("alloc memory failed for %s\n", path);
 413                 return ERR_PTR(-ENOMEM);
 414         }
 415
 416         strcpy(obj->path, path);
 417         obj->efile.fd = -1;
 418
 419         /*
 420          * Caller of this function should also calls
 421          * bpf_object__elf_finish() after data collection to return
 422          * obj_buf to user. If not, we should duplicate the buffer to
 423          * avoid user freeing them before elf finish.
 424          */
 425         obj->efile.obj_buf = obj_buf;
 426         obj->efile.obj_buf_sz = obj_buf_sz;
 427         obj->efile.maps_shndx = -1;
 428
 429         obj->loaded = false;
 430
 431         INIT_LIST_HEAD(&obj->list);
 432         list_add(&obj->list, &bpf_objects_list);
 433         return obj;
 434 }
 435
 436 static void bpf_object__elf_finish(struct bpf_object *obj)
 437 {
 438         if (!obj_elf_valid(obj))
 439                 return;
 440
 441         if (obj->efile.elf) {
 442                 elf_end(obj->efile.elf);
 443                 obj->efile.elf = NULL;
 444         }
 445         obj->efile.symbols = NULL;
 446
 447         zfree(&obj->efile.reloc);
 448         obj->efile.nr_reloc = 0;
 449         zclose(obj->efile.fd);
 450         obj->efile.obj_buf = NULL;
 451         obj->efile.obj_buf_sz = 0;
 452 }
 453
 454 static int bpf_object__elf_init(struct bpf_object *obj)
 455 {
 456         int err = 0;
 457         GElf_Ehdr *ep;
 458
 459         if (obj_elf_valid(obj)) {
 460                 pr_warning("elf init: internal error\n");
 461                 return -LIBBPF_ERRNO__LIBELF;
 462         }
 463
 464         if (obj->efile.obj_buf_sz > 0) {
 465                 /*
 466                  * obj_buf should have been validated by
 467                  * bpf_object__open_buffer().
 468                  */
 469                 obj->efile.elf = elf_memory(obj->efile.obj_buf,
 470                                             obj->efile.obj_buf_sz);
 471         } else {
 472                 obj->efile.fd = open(obj->path, O_RDONLY);
 473                 if (obj->efile.fd < 0) {
 474                         char errmsg[STRERR_BUFSIZE];
 475                         char *cp = str_error(errno, errmsg, sizeof(errmsg));
 476
 477                         pr_warning("failed to open %s: %s\n", obj->path, cp);
 478                         return -errno;
 479                 }
 480
 481                 obj->efile.elf = elf_begin(obj->efile.fd,
 482                                 LIBBPF_ELF_C_READ_MMAP,
 483                                 NULL);
 484         }
 485
 486         if (!obj->efile.elf) {
 487                 pr_warning("failed to open %s as ELF file\n",
 488                                 obj->path);
 489                 err = -LIBBPF_ERRNO__LIBELF;
 490                 goto errout;
 491         }
 492
 493         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 494                 pr_warning("failed to get EHDR from %s\n",
 495                                 obj->path);
 496                 err = -LIBBPF_ERRNO__FORMAT;
 497                 goto errout;
 498         }
 499         ep = &obj->efile.ehdr;
 500
 501         /* Old LLVM set e_machine to EM_NONE */
 502         if ((ep->e_type != ET_REL) || (ep->e_machine && (ep->e_machine != EM_BPF))) {
 503                 pr_warning("%s is not an eBPF object file\n",
 504                         obj->path);
 505                 err = -LIBBPF_ERRNO__FORMAT;
 506                 goto errout;
 507         }
 508
 509         return 0;
 510 errout:
 511         bpf_object__elf_finish(obj);
 512         return err;
 513 }
 514
 515 static int
 516 bpf_object__check_endianness(struct bpf_object *obj)
 517 {
 518         static unsigned int const endian = 1;
 519
 520         switch (obj->efile.ehdr.e_ident[EI_DATA]) {
 521         case ELFDATA2LSB:
 522                 /* We are big endian, BPF obj is little endian. */
 523                 if (*(unsigned char const *)&endian != 1)
 524                         goto mismatch;
 525                 break;
 526
 527         case ELFDATA2MSB:
 528                 /* We are little endian, BPF obj is big endian. */
 529                 if (*(unsigned char const *)&endian != 0)
 530                         goto mismatch;
 531                 break;
 532         default:
 533                 return -LIBBPF_ERRNO__ENDIAN;
 534         }
 535
 536         return 0;
 537
 538 mismatch:
 539         pr_warning("Error: endianness mismatch.\n");
 540         return -LIBBPF_ERRNO__ENDIAN;
 541 }
 542
 543 static int
 544 bpf_object__init_license(struct bpf_object *obj,
 545                          void *data, size_t size)
 546 {
 547         memcpy(obj->license, data,
 548                min(size, sizeof(obj->license) - 1));
 549         pr_debug("license of %s is %s\n", obj->path, obj->license);
 550         return 0;
 551 }
 552
 553 static int
 554 bpf_object__init_kversion(struct bpf_object *obj,
 555                           void *data, size_t size)
 556 {
 557         u32 kver;
 558
 559         if (size != sizeof(kver)) {
 560                 pr_warning("invalid kver section in %s\n", obj->path);
 561                 return -LIBBPF_ERRNO__FORMAT;
 562         }
 563         memcpy(&kver, data, sizeof(kver));
 564         obj->kern_version = kver;
 565         pr_debug("kernel version of %s is %x\n", obj->path,
 566                  obj->kern_version);
 567         return 0;
 568 }
 569
 570 static int compare_bpf_map(const void *_a, const void *_b)
 571 {
 572         const struct bpf_map *a = _a;
 573         const struct bpf_map *b = _b;
 574
 575         return a->offset - b->offset;
 576 }
 577
 578 static int
 579 bpf_object__init_maps(struct bpf_object *obj)
 580 {
 581         int i, map_idx, map_def_sz, nr_maps = 0;
 582         Elf_Scn *scn;
 583         Elf_Data *data;
 584         Elf_Data *symbols = obj->efile.symbols;
 585
 586         if (obj->efile.maps_shndx < 0)
 587                 return -EINVAL;
 588         if (!symbols)
 589                 return -EINVAL;
 590
 591         scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
 592         if (scn)
 593                 data = elf_getdata(scn, NULL);
 594         if (!scn || !data) {
 595                 pr_warning("failed to get Elf_Data from map section %d\n",
 596                            obj->efile.maps_shndx);
 597                 return -EINVAL;
 598         }
 599
 600         /*
 601          * Count number of maps. Each map has a name.
 602          * Array of maps is not supported: only the first element is
 603          * considered.
 604          *
 605          * TODO: Detect array of map and report error.
 606          */
 607         for (i = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 608                 GElf_Sym sym;
 609
 610                 if (!gelf_getsym(symbols, i, &sym))
 611                         continue;
 612                 if (sym.st_shndx != obj->efile.maps_shndx)
 613                         continue;
 614                 nr_maps++;
 615         }
 616
 617         /* Alloc obj->maps and fill nr_maps. */
 618         pr_debug("maps in %s: %d maps in %zd bytes\n", obj->path,
 619                  nr_maps, data->d_size);
 620
 621         if (!nr_maps)
 622                 return 0;
 623
 624         /* Assume equally sized map definitions */
 625         map_def_sz = data->d_size / nr_maps;
 626         if (!data->d_size || (data->d_size % nr_maps) != 0) {
 627                 pr_warning("unable to determine map definition size "
 628                            "section %s, %d maps in %zd bytes\n",
 629                            obj->path, nr_maps, data->d_size);
 630                 return -EINVAL;
 631         }
 632
 633         obj->maps = calloc(nr_maps, sizeof(obj->maps[0]));
 634         if (!obj->maps) {
 635                 pr_warning("alloc maps for object failed\n");
 636                 return -ENOMEM;
 637         }
 638         obj->nr_maps = nr_maps;
 639
 640         /*
 641          * fill all fd with -1 so won't close incorrect
 642          * fd (fd=0 is stdin) when failure (zclose won't close
 643          * negative fd)).
 644          */
 645         for (i = 0; i < nr_maps; i++)
 646                 obj->maps[i].fd = -1;
 647
 648         /*
 649          * Fill obj->maps using data in "maps" section.
 650          */
 651         for (i = 0, map_idx = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 652                 GElf_Sym sym;
 653                 const char *map_name;
 654                 struct bpf_map_def *def;
 655
 656                 if (!gelf_getsym(symbols, i, &sym))
 657                         continue;
 658                 if (sym.st_shndx != obj->efile.maps_shndx)
 659                         continue;
 660
 661                 map_name = elf_strptr(obj->efile.elf,
 662                                       obj->efile.strtabidx,
 663                                       sym.st_name);
 664                 obj->maps[map_idx].offset = sym.st_value;
 665                 if (sym.st_value + map_def_sz > data->d_size) {
 666                         pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
 667                                    obj->path, map_name);
 668                         return -EINVAL;
 669                 }
 670
 671                 obj->maps[map_idx].name = strdup(map_name);
 672                 if (!obj->maps[map_idx].name) {
 673                         pr_warning("failed to alloc map name\n");
 674                         return -ENOMEM;
 675                 }
 676                 pr_debug("map %d is \"%s\"\n", map_idx,
 677                          obj->maps[map_idx].name);
 678                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
 679                 /*
 680                  * If the definition of the map in the object file fits in
 681                  * bpf_map_def, copy it.  Any extra fields in our version
 682                  * of bpf_map_def will default to zero as a result of the
 683                  * calloc above.
 684                  */
 685                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
 686                         memcpy(&obj->maps[map_idx].def, def, map_def_sz);
 687                 } else {
 688                         /*
 689                          * Here the map structure being read is bigger than what
 690                          * we expect, truncate if the excess bits are all zero.
 691                          * If they are not zero, reject this map as
 692                          * incompatible.
 693                          */
 694                         char *b;
 695                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
 696                              b < ((char *)def) + map_def_sz; b++) {
 697                                 if (*b != 0) {
 698                                         pr_warning("maps section in %s: \"%s\" "
 699                                                    "has unrecognized, non-zero "
 700                                                    "options\n",
 701                                                    obj->path, map_name);
 702                                         return -EINVAL;
 703                                 }
 704                         }
 705                         memcpy(&obj->maps[map_idx].def, def,
 706                                sizeof(struct bpf_map_def));
 707                 }
 708                 map_idx++;
 709         }
 710
 711         qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]), compare_bpf_map);
 712         return 0;
 713 }
 714
 715 static bool section_have_execinstr(struct bpf_object *obj, int idx)
 716 {
 717         Elf_Scn *scn;
 718         GElf_Shdr sh;
 719
 720         scn = elf_getscn(obj->efile.elf, idx);
 721         if (!scn)
 722                 return false;
 723
 724         if (gelf_getshdr(scn, &sh) != &sh)
 725                 return false;
 726
 727         if (sh.sh_flags & SHF_EXECINSTR)
 728                 return true;
 729
 730         return false;
 731 }
 732
 733 static int bpf_object__elf_collect(struct bpf_object *obj)
 734 {
 735         Elf *elf = obj->efile.elf;
 736         GElf_Ehdr *ep = &obj->efile.ehdr;
 737         Elf_Scn *scn = NULL;
 738         int idx = 0, err = 0;
 739
 740         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
 741         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
 742                 pr_warning("failed to get e_shstrndx from %s\n",
 743                            obj->path);
 744                 return -LIBBPF_ERRNO__FORMAT;
 745         }
 746
 747         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 748                 char *name;
 749                 GElf_Shdr sh;
 750                 Elf_Data *data;
 751
 752                 idx++;
 753                 if (gelf_getshdr(scn, &sh) != &sh) {
 754                         pr_warning("failed to get section(%d) header from %s\n",
 755                                    idx, obj->path);
 756                         err = -LIBBPF_ERRNO__FORMAT;
 757                         goto out;
 758                 }
 759
 760                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 761                 if (!name) {
 762                         pr_warning("failed to get section(%d) name from %s\n",
 763                                    idx, obj->path);
 764                         err = -LIBBPF_ERRNO__FORMAT;
 765                         goto out;
 766                 }
 767
 768                 data = elf_getdata(scn, 0);
 769                 if (!data) {
 770                         pr_warning("failed to get section(%d) data from %s(%s)\n",
 771                                    idx, name, obj->path);
 772                         err = -LIBBPF_ERRNO__FORMAT;
 773                         goto out;
 774                 }
 775                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
 776                          idx, name, (unsigned long)data->d_size,
 777                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
 778                          (int)sh.sh_type);
 779
 780                 if (strcmp(name, "license") == 0)
 781                         err = bpf_object__init_license(obj,
 782                                                        data->d_buf,
 783                                                        data->d_size);
 784                 else if (strcmp(name, "version") == 0)
 785                         err = bpf_object__init_kversion(obj,
 786                                                         data->d_buf,
 787                                                         data->d_size);
 788                 else if (strcmp(name, "maps") == 0)
 789                         obj->efile.maps_shndx = idx;
 790                 else if (strcmp(name, BTF_ELF_SEC) == 0) {
 791                         obj->btf = btf__new(data->d_buf, data->d_size,
 792                                             __pr_debug);
 793                         if (IS_ERR(obj->btf)) {
 794                                 pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
 795                                            BTF_ELF_SEC, PTR_ERR(obj->btf));
 796                                 obj->btf = NULL;
 797                         }
 798                 } else if (sh.sh_type == SHT_SYMTAB) {
 799                         if (obj->efile.symbols) {
 800                                 pr_warning("bpf: multiple SYMTAB in %s\n",
 801                                            obj->path);
 802                                 err = -LIBBPF_ERRNO__FORMAT;
 803                         } else {
 804                                 obj->efile.symbols = data;
 805                                 obj->efile.strtabidx = sh.sh_link;
 806                         }
 807                 } else if ((sh.sh_type == SHT_PROGBITS) &&
 808                            (sh.sh_flags & SHF_EXECINSTR) &&
 809                            (data->d_size > 0)) {
 810                         if (strcmp(name, ".text") == 0)
 811                                 obj->efile.text_shndx = idx;
 812                         err = bpf_object__add_program(obj, data->d_buf,
 813                                                       data->d_size, name, idx);
 814                         if (err) {
 815                                 char errmsg[STRERR_BUFSIZE];
 816                                 char *cp = str_error(-err, errmsg, sizeof(errmsg));
 817
 818                                 pr_warning("failed to alloc program %s (%s): %s",
 819                                            name, obj->path, cp);
 820                         }
 821                 } else if (sh.sh_type == SHT_REL) {
 822                         void *reloc = obj->efile.reloc;
 823                         int nr_reloc = obj->efile.nr_reloc + 1;
 824                         int sec = sh.sh_info; /* points to other section */
 825
 826                         /* Only do relo for section with exec instructions */
 827                         if (!section_have_execinstr(obj, sec)) {
 828                                 pr_debug("skip relo %s(%d) for section(%d)\n",
 829                                          name, idx, sec);
 830                                 continue;
 831                         }
 832
 833                         reloc = reallocarray(reloc, nr_reloc,
 834                                              sizeof(*obj->efile.reloc));
 835                         if (!reloc) {
 836                                 pr_warning("realloc failed\n");
 837                                 err = -ENOMEM;
 838                         } else {
 839                                 int n = nr_reloc - 1;
 840
 841                                 obj->efile.reloc = reloc;
 842                                 obj->efile.nr_reloc = nr_reloc;
 843
 844                                 obj->efile.reloc[n].shdr = sh;
 845                                 obj->efile.reloc[n].data = data;
 846                         }
 847                 } else {
 848                         pr_debug("skip section(%d) %s\n", idx, name);
 849                 }
 850                 if (err)
 851                         goto out;
 852         }
 853
 854         if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
 855                 pr_warning("Corrupted ELF file: index of strtab invalid\n");
 856                 return LIBBPF_ERRNO__FORMAT;
 857         }
 858         if (obj->efile.maps_shndx >= 0) {
 859                 err = bpf_object__init_maps(obj);
 860                 if (err)
 861                         goto out;
 862         }
 863         err = bpf_object__init_prog_names(obj);
 864 out:
 865         return err;
 866 }
 867
 868 static struct bpf_program *
 869 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
 870 {
 871         struct bpf_program *prog;
 872         size_t i;
 873
 874         for (i = 0; i < obj->nr_programs; i++) {
 875                 prog = &obj->programs[i];
 876                 if (prog->idx == idx)
 877                         return prog;
 878         }
 879         return NULL;
 880 }
 881
 882 struct bpf_program *
 883 bpf_object__find_program_by_title(struct bpf_object *obj, const char *title)
 884 {
 885         struct bpf_program *pos;
 886
 887         bpf_object__for_each_program(pos, obj) {
 888                 if (pos->section_name && !strcmp(pos->section_name, title))
 889                         return pos;
 890         }
 891         return NULL;
 892 }
 893
 894 static int
 895 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
 896                            Elf_Data *data, struct bpf_object *obj)
 897 {
 898         Elf_Data *symbols = obj->efile.symbols;
 899         int text_shndx = obj->efile.text_shndx;
 900         int maps_shndx = obj->efile.maps_shndx;
 901         struct bpf_map *maps = obj->maps;
 902         size_t nr_maps = obj->nr_maps;
 903         int i, nrels;
 904
 905         pr_debug("collecting relocating info for: '%s'\n",
 906                  prog->section_name);
 907         nrels = shdr->sh_size / shdr->sh_entsize;
 908
 909         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
 910         if (!prog->reloc_desc) {
 911                 pr_warning("failed to alloc memory in relocation\n");
 912                 return -ENOMEM;
 913         }
 914         prog->nr_reloc = nrels;
 915
 916         for (i = 0; i < nrels; i++) {
 917                 GElf_Sym sym;
 918                 GElf_Rel rel;
 919                 unsigned int insn_idx;
 920                 struct bpf_insn *insns = prog->insns;
 921                 size_t map_idx;
 922
 923                 if (!gelf_getrel(data, i, &rel)) {
 924                         pr_warning("relocation: failed to get %d reloc\n", i);
 925                         return -LIBBPF_ERRNO__FORMAT;
 926                 }
 927
 928                 if (!gelf_getsym(symbols,
 929                                  GELF_R_SYM(rel.r_info),
 930                                  &sym)) {
 931                         pr_warning("relocation: symbol %"PRIx64" not found\n",
 932                                    GELF_R_SYM(rel.r_info));
 933                         return -LIBBPF_ERRNO__FORMAT;
 934                 }
 935                 pr_debug("relo for %lld value %lld name %d\n",
 936                          (long long) (rel.r_info >> 32),
 937                          (long long) sym.st_value, sym.st_name);
 938
 939                 if (sym.st_shndx != maps_shndx && sym.st_shndx != text_shndx) {
 940                         pr_warning("Program '%s' contains non-map related relo data pointing to section %u\n",
 941                                    prog->section_name, sym.st_shndx);
 942                         return -LIBBPF_ERRNO__RELOC;
 943                 }
 944
 945                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
 946                 pr_debug("relocation: insn_idx=%u\n", insn_idx);
 947
 948                 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
 949                         if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
 950                                 pr_warning("incorrect bpf_call opcode\n");
 951                                 return -LIBBPF_ERRNO__RELOC;
 952                         }
 953                         prog->reloc_desc[i].type = RELO_CALL;
 954                         prog->reloc_desc[i].insn_idx = insn_idx;
 955                         prog->reloc_desc[i].text_off = sym.st_value;
 956                         obj->has_pseudo_calls = true;
 957                         continue;
 958                 }
 959
 960                 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
 961                         pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
 962                                    insn_idx, insns[insn_idx].code);
 963                         return -LIBBPF_ERRNO__RELOC;
 964                 }
 965
 966                 /* TODO: 'maps' is sorted. We can use bsearch to make it faster. */
 967                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
 968                         if (maps[map_idx].offset == sym.st_value) {
 969                                 pr_debug("relocation: find map %zd (%s) for insn %u\n",
 970                                          map_idx, maps[map_idx].name, insn_idx);
 971                                 break;
 972                         }
 973                 }
 974
 975                 if (map_idx >= nr_maps) {
 976                         pr_warning("bpf relocation: map_idx %d large than %d\n",
 977                                    (int)map_idx, (int)nr_maps - 1);
 978                         return -LIBBPF_ERRNO__RELOC;
 979                 }
 980
 981                 prog->reloc_desc[i].type = RELO_LD64;
 982                 prog->reloc_desc[i].insn_idx = insn_idx;
 983                 prog->reloc_desc[i].map_idx = map_idx;
 984         }
 985         return 0;
 986 }
 987
 988 static int bpf_map_find_btf_info(struct bpf_map *map, const struct btf *btf)
 989 {
 990         const struct btf_type *container_type;
 991         const struct btf_member *key, *value;
 992         struct bpf_map_def *def = &map->def;
 993         const size_t max_name = 256;
 994         char container_name[max_name];
 995         __s64 key_size, value_size;
 996         __s32 container_id;
 997
 998         if (snprintf(container_name, max_name, "____btf_map_%s", map->name) ==
 999             max_name) {
1000                 pr_warning("map:%s length of '____btf_map_%s' is too long\n",
1001                            map->name, map->name);
1002                 return -EINVAL;
1003         }
1004
1005         container_id = btf__find_by_name(btf, container_name);
1006         if (container_id < 0) {
1007                 pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
1008                          map->name, container_name);
1009                 return container_id;
1010         }
1011
1012         container_type = btf__type_by_id(btf, container_id);
1013         if (!container_type) {
1014                 pr_warning("map:%s cannot find BTF type for container_id:%u\n",
1015                            map->name, container_id);
1016                 return -EINVAL;
1017         }
1018
1019         if (BTF_INFO_KIND(container_type->info) != BTF_KIND_STRUCT ||
1020             BTF_INFO_VLEN(container_type->info) < 2) {
1021                 pr_warning("map:%s container_name:%s is an invalid container struct\n",
1022                            map->name, container_name);
1023                 return -EINVAL;
1024         }
1025
1026         key = (struct btf_member *)(container_type + 1);
1027         value = key + 1;
1028
1029         key_size = btf__resolve_size(btf, key->type);
1030         if (key_size < 0) {
1031                 pr_warning("map:%s invalid BTF key_type_size\n",
1032                            map->name);
1033                 return key_size;
1034         }
1035
1036         if (def->key_size != key_size) {
1037                 pr_warning("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
1038                            map->name, (__u32)key_size, def->key_size);
1039                 return -EINVAL;
1040         }
1041
1042         value_size = btf__resolve_size(btf, value->type);
1043         if (value_size < 0) {
1044                 pr_warning("map:%s invalid BTF value_type_size\n", map->name);
1045                 return value_size;
1046         }
1047
1048         if (def->value_size != value_size) {
1049                 pr_warning("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
1050                            map->name, (__u32)value_size, def->value_size);
1051                 return -EINVAL;
1052         }
1053
1054         map->btf_key_type_id = key->type;
1055         map->btf_value_type_id = value->type;
1056
1057         return 0;
1058 }
1059
1060 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
1061 {
1062         struct bpf_map_info info = {};
1063         __u32 len = sizeof(info);
1064         int new_fd, err;
1065         char *new_name;
1066
1067         err = bpf_obj_get_info_by_fd(fd, &info, &len);
1068         if (err)
1069                 return err;
1070
1071         new_name = strdup(info.name);
1072         if (!new_name)
1073                 return -errno;
1074
1075         new_fd = open("/", O_RDONLY | O_CLOEXEC);
1076         if (new_fd < 0)
1077                 goto err_free_new_name;
1078
1079         new_fd = dup3(fd, new_fd, O_CLOEXEC);
1080         if (new_fd < 0)
1081                 goto err_close_new_fd;
1082
1083         err = zclose(map->fd);
1084         if (err)
1085                 goto err_close_new_fd;
1086         free(map->name);
1087
1088         map->fd = new_fd;
1089         map->name = new_name;
1090         map->def.type = info.type;
1091         map->def.key_size = info.key_size;
1092         map->def.value_size = info.value_size;
1093         map->def.max_entries = info.max_entries;
1094         map->def.map_flags = info.map_flags;
1095         map->btf_key_type_id = info.btf_key_type_id;
1096         map->btf_value_type_id = info.btf_value_type_id;
1097
1098         return 0;
1099
1100 err_close_new_fd:
1101         close(new_fd);
1102 err_free_new_name:
1103         free(new_name);
1104         return -errno;
1105 }
1106
1107 static int
1108 bpf_object__create_maps(struct bpf_object *obj)
1109 {
1110         struct bpf_create_map_attr create_attr = {};
1111         unsigned int i;
1112         int err;
1113
1114         for (i = 0; i < obj->nr_maps; i++) {
1115                 struct bpf_map *map = &obj->maps[i];
1116                 struct bpf_map_def *def = &map->def;
1117                 char *cp, errmsg[STRERR_BUFSIZE];
1118                 int *pfd = &map->fd;
1119
1120                 if (map->fd >= 0) {
1121                         pr_debug("skip map create (preset) %s: fd=%d\n",
1122                                  map->name, map->fd);
1123                         continue;
1124                 }
1125
1126                 create_attr.name = map->name;
1127                 create_attr.map_ifindex = map->map_ifindex;
1128                 create_attr.map_type = def->type;
1129                 create_attr.map_flags = def->map_flags;
1130                 create_attr.key_size = def->key_size;
1131                 create_attr.value_size = def->value_size;
1132                 create_attr.max_entries = def->max_entries;
1133                 create_attr.btf_fd = 0;
1134                 create_attr.btf_key_type_id = 0;
1135                 create_attr.btf_value_type_id = 0;
1136
1137                 if (obj->btf && !bpf_map_find_btf_info(map, obj->btf)) {
1138                         create_attr.btf_fd = btf__fd(obj->btf);
1139                         create_attr.btf_key_type_id = map->btf_key_type_id;
1140                         create_attr.btf_value_type_id = map->btf_value_type_id;
1141                 }
1142
1143                 *pfd = bpf_create_map_xattr(&create_attr);
1144                 if (*pfd < 0 && create_attr.btf_key_type_id) {
1145                         cp = str_error(errno, errmsg, sizeof(errmsg));
1146                         pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
1147                                    map->name, cp, errno);
1148                         create_attr.btf_fd = 0;
1149                         create_attr.btf_key_type_id = 0;
1150                         create_attr.btf_value_type_id = 0;
1151                         map->btf_key_type_id = 0;
1152                         map->btf_value_type_id = 0;
1153                         *pfd = bpf_create_map_xattr(&create_attr);
1154                 }
1155
1156                 if (*pfd < 0) {
1157                         size_t j;
1158
1159                         err = *pfd;
1160                         cp = str_error(errno, errmsg, sizeof(errmsg));
1161                         pr_warning("failed to create map (name: '%s'): %s\n",
1162                                    map->name, cp);
1163                         for (j = 0; j < i; j++)
1164                                 zclose(obj->maps[j].fd);
1165                         return err;
1166                 }
1167                 pr_debug("create map %s: fd=%d\n", map->name, *pfd);
1168         }
1169
1170         return 0;
1171 }
1172
1173 static int
1174 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
1175                         struct reloc_desc *relo)
1176 {
1177         struct bpf_insn *insn, *new_insn;
1178         struct bpf_program *text;
1179         size_t new_cnt;
1180
1181         if (relo->type != RELO_CALL)
1182                 return -LIBBPF_ERRNO__RELOC;
1183
1184         if (prog->idx == obj->efile.text_shndx) {
1185                 pr_warning("relo in .text insn %d into off %d\n",
1186                            relo->insn_idx, relo->text_off);
1187                 return -LIBBPF_ERRNO__RELOC;
1188         }
1189
1190         if (prog->main_prog_cnt == 0) {
1191                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
1192                 if (!text) {
1193                         pr_warning("no .text section found yet relo into text exist\n");
1194                         return -LIBBPF_ERRNO__RELOC;
1195                 }
1196                 new_cnt = prog->insns_cnt + text->insns_cnt;
1197                 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
1198                 if (!new_insn) {
1199                         pr_warning("oom in prog realloc\n");
1200                         return -ENOMEM;
1201                 }
1202                 memcpy(new_insn + prog->insns_cnt, text->insns,
1203                        text->insns_cnt * sizeof(*insn));
1204                 prog->insns = new_insn;
1205                 prog->main_prog_cnt = prog->insns_cnt;
1206                 prog->insns_cnt = new_cnt;
1207                 pr_debug("added %zd insn from %s to prog %s\n",
1208                          text->insns_cnt, text->section_name,
1209                          prog->section_name);
1210         }
1211         insn = &prog->insns[relo->insn_idx];
1212         insn->imm += prog->main_prog_cnt - relo->insn_idx;
1213         return 0;
1214 }
1215
1216 static int
1217 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
1218 {
1219         int i, err;
1220
1221         if (!prog || !prog->reloc_desc)
1222                 return 0;
1223
1224         for (i = 0; i < prog->nr_reloc; i++) {
1225                 if (prog->reloc_desc[i].type == RELO_LD64) {
1226                         struct bpf_insn *insns = prog->insns;
1227                         int insn_idx, map_idx;
1228
1229                         insn_idx = prog->reloc_desc[i].insn_idx;
1230                         map_idx = prog->reloc_desc[i].map_idx;
1231
1232                         if (insn_idx >= (int)prog->insns_cnt) {
1233                                 pr_warning("relocation out of range: '%s'\n",
1234                                            prog->section_name);
1235                                 return -LIBBPF_ERRNO__RELOC;
1236                         }
1237                         insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
1238                         insns[insn_idx].imm = obj->maps[map_idx].fd;
1239                 } else {
1240                         err = bpf_program__reloc_text(prog, obj,
1241                                                       &prog->reloc_desc[i]);
1242                         if (err)
1243                                 return err;
1244                 }
1245         }
1246
1247         zfree(&prog->reloc_desc);
1248         prog->nr_reloc = 0;
1249         return 0;
1250 }
1251
1252
1253 static int
1254 bpf_object__relocate(struct bpf_object *obj)
1255 {
1256         struct bpf_program *prog;
1257         size_t i;
1258         int err;
1259
1260         for (i = 0; i < obj->nr_programs; i++) {
1261                 prog = &obj->programs[i];
1262
1263                 err = bpf_program__relocate(prog, obj);
1264                 if (err) {
1265                         pr_warning("failed to relocate '%s'\n",
1266                                    prog->section_name);
1267                         return err;
1268                 }
1269         }
1270         return 0;
1271 }
1272
1273 static int bpf_object__collect_reloc(struct bpf_object *obj)
1274 {
1275         int i, err;
1276
1277         if (!obj_elf_valid(obj)) {
1278                 pr_warning("Internal error: elf object is closed\n");
1279                 return -LIBBPF_ERRNO__INTERNAL;
1280         }
1281
1282         for (i = 0; i < obj->efile.nr_reloc; i++) {
1283                 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
1284                 Elf_Data *data = obj->efile.reloc[i].data;
1285                 int idx = shdr->sh_info;
1286                 struct bpf_program *prog;
1287
1288                 if (shdr->sh_type != SHT_REL) {
1289                         pr_warning("internal error at %d\n", __LINE__);
1290                         return -LIBBPF_ERRNO__INTERNAL;
1291                 }
1292
1293                 prog = bpf_object__find_prog_by_idx(obj, idx);
1294                 if (!prog) {
1295                         pr_warning("relocation failed: no section(%d)\n", idx);
1296                         return -LIBBPF_ERRNO__RELOC;
1297                 }
1298
1299                 err = bpf_program__collect_reloc(prog,
1300                                                  shdr, data,
1301                                                  obj);
1302                 if (err)
1303                         return err;
1304         }
1305         return 0;
1306 }
1307
1308 static int
1309 load_program(enum bpf_prog_type type, enum bpf_attach_type expected_attach_type,
1310              const char *name, struct bpf_insn *insns, int insns_cnt,
1311              char *license, u32 kern_version, int *pfd, int prog_ifindex)
1312 {
1313         struct bpf_load_program_attr load_attr;
1314         char *cp, errmsg[STRERR_BUFSIZE];
1315         char *log_buf;
1316         int ret;
1317
1318         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
1319         load_attr.prog_type = type;
1320         load_attr.expected_attach_type = expected_attach_type;
1321         load_attr.name = name;
1322         load_attr.insns = insns;
1323         load_attr.insns_cnt = insns_cnt;
1324         load_attr.license = license;
1325         load_attr.kern_version = kern_version;
1326         load_attr.prog_ifindex = prog_ifindex;
1327
1328         if (!load_attr.insns || !load_attr.insns_cnt)
1329                 return -EINVAL;
1330
1331         log_buf = malloc(BPF_LOG_BUF_SIZE);
1332         if (!log_buf)
1333                 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
1334
1335         ret = bpf_load_program_xattr(&load_attr, log_buf, BPF_LOG_BUF_SIZE);
1336
1337         if (ret >= 0) {
1338                 *pfd = ret;
1339                 ret = 0;
1340                 goto out;
1341         }
1342
1343         ret = -LIBBPF_ERRNO__LOAD;
1344         cp = str_error(errno, errmsg, sizeof(errmsg));
1345         pr_warning("load bpf program failed: %s\n", cp);
1346
1347         if (log_buf && log_buf[0] != '\0') {
1348                 ret = -LIBBPF_ERRNO__VERIFY;
1349                 pr_warning("-- BEGIN DUMP LOG ---\n");
1350                 pr_warning("\n%s\n", log_buf);
1351                 pr_warning("-- END LOG --\n");
1352         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
1353                 pr_warning("Program too large (%zu insns), at most %d insns\n",
1354                            load_attr.insns_cnt, BPF_MAXINSNS);
1355                 ret = -LIBBPF_ERRNO__PROG2BIG;
1356         } else {
1357                 /* Wrong program type? */
1358                 if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
1359                         int fd;
1360
1361                         load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
1362                         load_attr.expected_attach_type = 0;
1363                         fd = bpf_load_program_xattr(&load_attr, NULL, 0);
1364                         if (fd >= 0) {
1365                                 close(fd);
1366                                 ret = -LIBBPF_ERRNO__PROGTYPE;
1367                                 goto out;
1368                         }
1369                 }
1370
1371                 if (log_buf)
1372                         ret = -LIBBPF_ERRNO__KVER;
1373         }
1374
1375 out:
1376         free(log_buf);
1377         return ret;
1378 }
1379
1380 static int
1381 bpf_program__load(struct bpf_program *prog,
1382                   char *license, u32 kern_version)
1383 {
1384         int err = 0, fd, i;
1385
1386         if (prog->instances.nr < 0 || !prog->instances.fds) {
1387                 if (prog->preprocessor) {
1388                         pr_warning("Internal error: can't load program '%s'\n",
1389                                    prog->section_name);
1390                         return -LIBBPF_ERRNO__INTERNAL;
1391                 }
1392
1393                 prog->instances.fds = malloc(sizeof(int));
1394                 if (!prog->instances.fds) {
1395                         pr_warning("Not enough memory for BPF fds\n");
1396                         return -ENOMEM;
1397                 }
1398                 prog->instances.nr = 1;
1399                 prog->instances.fds[0] = -1;
1400         }
1401
1402         if (!prog->preprocessor) {
1403                 if (prog->instances.nr != 1) {
1404                         pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
1405                                    prog->section_name, prog->instances.nr);
1406                 }
1407                 err = load_program(prog->type, prog->expected_attach_type,
1408                                    prog->name, prog->insns, prog->insns_cnt,
1409                                    license, kern_version, &fd,
1410                                    prog->prog_ifindex);
1411                 if (!err)
1412                         prog->instances.fds[0] = fd;
1413                 goto out;
1414         }
1415
1416         for (i = 0; i < prog->instances.nr; i++) {
1417                 struct bpf_prog_prep_result result;
1418                 bpf_program_prep_t preprocessor = prog->preprocessor;
1419
1420                 bzero(&result, sizeof(result));
1421                 err = preprocessor(prog, i, prog->insns,
1422                                    prog->insns_cnt, &result);
1423                 if (err) {
1424                         pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
1425                                    i, prog->section_name);
1426                         goto out;
1427                 }
1428
1429                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
1430                         pr_debug("Skip loading the %dth instance of program '%s'\n",
1431                                  i, prog->section_name);
1432                         prog->instances.fds[i] = -1;
1433                         if (result.pfd)
1434                                 *result.pfd = -1;
1435                         continue;
1436                 }
1437
1438                 err = load_program(prog->type, prog->expected_attach_type,
1439                                    prog->name, result.new_insn_ptr,
1440                                    result.new_insn_cnt,
1441                                    license, kern_version, &fd,
1442                                    prog->prog_ifindex);
1443
1444                 if (err) {
1445                         pr_warning("Loading the %dth instance of program '%s' failed\n",
1446                                         i, prog->section_name);
1447                         goto out;
1448                 }
1449
1450                 if (result.pfd)
1451                         *result.pfd = fd;
1452                 prog->instances.fds[i] = fd;
1453         }
1454 out:
1455         if (err)
1456                 pr_warning("failed to load program '%s'\n",
1457                            prog->section_name);
1458         zfree(&prog->insns);
1459         prog->insns_cnt = 0;
1460         return err;
1461 }
1462
1463 static bool bpf_program__is_function_storage(struct bpf_program *prog,
1464                                              struct bpf_object *obj)
1465 {
1466         return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
1467 }
1468
1469 static int
1470 bpf_object__load_progs(struct bpf_object *obj)
1471 {
1472         size_t i;
1473         int err;
1474
1475         for (i = 0; i < obj->nr_programs; i++) {
1476                 if (bpf_program__is_function_storage(&obj->programs[i], obj))
1477                         continue;
1478                 err = bpf_program__load(&obj->programs[i],
1479                                         obj->license,
1480                                         obj->kern_version);
1481                 if (err)
1482                         return err;
1483         }
1484         return 0;
1485 }
1486
1487 static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
1488 {
1489         switch (type) {
1490         case BPF_PROG_TYPE_SOCKET_FILTER:
1491         case BPF_PROG_TYPE_SCHED_CLS:
1492         case BPF_PROG_TYPE_SCHED_ACT:
1493         case BPF_PROG_TYPE_XDP:
1494         case BPF_PROG_TYPE_CGROUP_SKB:
1495         case BPF_PROG_TYPE_CGROUP_SOCK:
1496         case BPF_PROG_TYPE_LWT_IN:
1497         case BPF_PROG_TYPE_LWT_OUT:
1498         case BPF_PROG_TYPE_LWT_XMIT:
1499         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
1500         case BPF_PROG_TYPE_SOCK_OPS:
1501         case BPF_PROG_TYPE_SK_SKB:
1502         case BPF_PROG_TYPE_CGROUP_DEVICE:
1503         case BPF_PROG_TYPE_SK_MSG:
1504         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1505         case BPF_PROG_TYPE_LIRC_MODE2:
1506         case BPF_PROG_TYPE_SK_REUSEPORT:
1507                 return false;
1508         case BPF_PROG_TYPE_UNSPEC:
1509         case BPF_PROG_TYPE_KPROBE:
1510         case BPF_PROG_TYPE_TRACEPOINT:
1511         case BPF_PROG_TYPE_PERF_EVENT:
1512         case BPF_PROG_TYPE_RAW_TRACEPOINT:
1513         default:
1514                 return true;
1515         }
1516 }
1517
1518 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
1519 {
1520         if (needs_kver && obj->kern_version == 0) {
1521                 pr_warning("%s doesn't provide kernel version\n",
1522                            obj->path);
1523                 return -LIBBPF_ERRNO__KVERSION;
1524         }
1525         return 0;
1526 }
1527
1528 static struct bpf_object *
1529 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
1530                    bool needs_kver)
1531 {
1532         struct bpf_object *obj;
1533         int err;
1534
1535         if (elf_version(EV_CURRENT) == EV_NONE) {
1536                 pr_warning("failed to init libelf for %s\n", path);
1537                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1538         }
1539
1540         obj = bpf_object__new(path, obj_buf, obj_buf_sz);
1541         if (IS_ERR(obj))
1542                 return obj;
1543
1544         CHECK_ERR(bpf_object__elf_init(obj), err, out);
1545         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
1546         CHECK_ERR(bpf_object__elf_collect(obj), err, out);
1547         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
1548         CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
1549
1550         bpf_object__elf_finish(obj);
1551         return obj;
1552 out:
1553         bpf_object__close(obj);
1554         return ERR_PTR(err);
1555 }
1556
1557 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
1558 {
1559         /* param validation */
1560         if (!attr->file)
1561                 return NULL;
1562
1563         pr_debug("loading %s\n", attr->file);
1564
1565         return __bpf_object__open(attr->file, NULL, 0,
1566                                   bpf_prog_type__needs_kver(attr->prog_type));
1567 }
1568
1569 struct bpf_object *bpf_object__open(const char *path)
1570 {
1571         struct bpf_object_open_attr attr = {
1572                 .file           = path,
1573                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
1574         };
1575
1576         return bpf_object__open_xattr(&attr);
1577 }
1578
1579 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
1580                                            size_t obj_buf_sz,
1581                                            const char *name)
1582 {
1583         char tmp_name[64];
1584
1585         /* param validation */
1586         if (!obj_buf || obj_buf_sz <= 0)
1587                 return NULL;
1588
1589         if (!name) {
1590                 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
1591                          (unsigned long)obj_buf,
1592                          (unsigned long)obj_buf_sz);
1593                 tmp_name[sizeof(tmp_name) - 1] = '\0';
1594                 name = tmp_name;
1595         }
1596         pr_debug("loading object '%s' from buffer\n",
1597                  name);
1598
1599         return __bpf_object__open(name, obj_buf, obj_buf_sz, true);
1600 }
1601
1602 int bpf_object__unload(struct bpf_object *obj)
1603 {
1604         size_t i;
1605
1606         if (!obj)
1607                 return -EINVAL;
1608
1609         for (i = 0; i < obj->nr_maps; i++)
1610                 zclose(obj->maps[i].fd);
1611
1612         for (i = 0; i < obj->nr_programs; i++)
1613                 bpf_program__unload(&obj->programs[i]);
1614
1615         return 0;
1616 }
1617
1618 int bpf_object__load(struct bpf_object *obj)
1619 {
1620         int err;
1621
1622         if (!obj)
1623                 return -EINVAL;
1624
1625         if (obj->loaded) {
1626                 pr_warning("object should not be loaded twice\n");
1627                 return -EINVAL;
1628         }
1629
1630         obj->loaded = true;
1631
1632         CHECK_ERR(bpf_object__create_maps(obj), err, out);
1633         CHECK_ERR(bpf_object__relocate(obj), err, out);
1634         CHECK_ERR(bpf_object__load_progs(obj), err, out);
1635
1636         return 0;
1637 out:
1638         bpf_object__unload(obj);
1639         pr_warning("failed to load object '%s'\n", obj->path);
1640         return err;
1641 }
1642
1643 static int check_path(const char *path)
1644 {
1645         char *cp, errmsg[STRERR_BUFSIZE];
1646         struct statfs st_fs;
1647         char *dname, *dir;
1648         int err = 0;
1649
1650         if (path == NULL)
1651                 return -EINVAL;
1652
1653         dname = strdup(path);
1654         if (dname == NULL)
1655                 return -ENOMEM;
1656
1657         dir = dirname(dname);
1658         if (statfs(dir, &st_fs)) {
1659                 cp = str_error(errno, errmsg, sizeof(errmsg));
1660                 pr_warning("failed to statfs %s: %s\n", dir, cp);
1661                 err = -errno;
1662         }
1663         free(dname);
1664
1665         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
1666                 pr_warning("specified path %s is not on BPF FS\n", path);
1667                 err = -EINVAL;
1668         }
1669
1670         return err;
1671 }
1672
1673 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
1674                               int instance)
1675 {
1676         char *cp, errmsg[STRERR_BUFSIZE];
1677         int err;
1678
1679         err = check_path(path);
1680         if (err)
1681                 return err;
1682
1683         if (prog == NULL) {
1684                 pr_warning("invalid program pointer\n");
1685                 return -EINVAL;
1686         }
1687
1688         if (instance < 0 || instance >= prog->instances.nr) {
1689                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
1690                            instance, prog->section_name, prog->instances.nr);
1691                 return -EINVAL;
1692         }
1693
1694         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
1695                 cp = str_error(errno, errmsg, sizeof(errmsg));
1696                 pr_warning("failed to pin program: %s\n", cp);
1697                 return -errno;
1698         }
1699         pr_debug("pinned program '%s'\n", path);
1700
1701         return 0;
1702 }
1703
1704 static int make_dir(const char *path)
1705 {
1706         char *cp, errmsg[STRERR_BUFSIZE];
1707         int err = 0;
1708
1709         if (mkdir(path, 0700) && errno != EEXIST)
1710                 err = -errno;
1711
1712         if (err) {
1713                 cp = str_error(-err, errmsg, sizeof(errmsg));
1714                 pr_warning("failed to mkdir %s: %s\n", path, cp);
1715         }
1716         return err;
1717 }
1718
1719 int bpf_program__pin(struct bpf_program *prog, const char *path)
1720 {
1721         int i, err;
1722
1723         err = check_path(path);
1724         if (err)
1725                 return err;
1726
1727         if (prog == NULL) {
1728                 pr_warning("invalid program pointer\n");
1729                 return -EINVAL;
1730         }
1731
1732         if (prog->instances.nr <= 0) {
1733                 pr_warning("no instances of prog %s to pin\n",
1734                            prog->section_name);
1735                 return -EINVAL;
1736         }
1737
1738         err = make_dir(path);
1739         if (err)
1740                 return err;
1741
1742         for (i = 0; i < prog->instances.nr; i++) {
1743                 char buf[PATH_MAX];
1744                 int len;
1745
1746                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
1747                 if (len < 0)
1748                         return -EINVAL;
1749                 else if (len >= PATH_MAX)
1750                         return -ENAMETOOLONG;
1751
1752                 err = bpf_program__pin_instance(prog, buf, i);
1753                 if (err)
1754                         return err;
1755         }
1756
1757         return 0;
1758 }
1759
1760 int bpf_map__pin(struct bpf_map *map, const char *path)
1761 {
1762         char *cp, errmsg[STRERR_BUFSIZE];
1763         int err;
1764
1765         err = check_path(path);
1766         if (err)
1767                 return err;
1768
1769         if (map == NULL) {
1770                 pr_warning("invalid map pointer\n");
1771                 return -EINVAL;
1772         }
1773
1774         if (bpf_obj_pin(map->fd, path)) {
1775                 cp = str_error(errno, errmsg, sizeof(errmsg));
1776                 pr_warning("failed to pin map: %s\n", cp);
1777                 return -errno;
1778         }
1779
1780         pr_debug("pinned map '%s'\n", path);
1781         return 0;
1782 }
1783
1784 int bpf_object__pin(struct bpf_object *obj, const char *path)
1785 {
1786         struct bpf_program *prog;
1787         struct bpf_map *map;
1788         int err;
1789
1790         if (!obj)
1791                 return -ENOENT;
1792
1793         if (!obj->loaded) {
1794                 pr_warning("object not yet loaded; load it first\n");
1795                 return -ENOENT;
1796         }
1797
1798         err = make_dir(path);
1799         if (err)
1800                 return err;
1801
1802         bpf_map__for_each(map, obj) {
1803                 char buf[PATH_MAX];
1804                 int len;
1805
1806                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1807                                bpf_map__name(map));
1808                 if (len < 0)
1809                         return -EINVAL;
1810                 else if (len >= PATH_MAX)
1811                         return -ENAMETOOLONG;
1812
1813                 err = bpf_map__pin(map, buf);
1814                 if (err)
1815                         return err;
1816         }
1817
1818         bpf_object__for_each_program(prog, obj) {
1819                 char buf[PATH_MAX];
1820                 int len;
1821
1822                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1823                                prog->section_name);
1824                 if (len < 0)
1825                         return -EINVAL;
1826                 else if (len >= PATH_MAX)
1827                         return -ENAMETOOLONG;
1828
1829                 err = bpf_program__pin(prog, buf);
1830                 if (err)
1831                         return err;
1832         }
1833
1834         return 0;
1835 }
1836
1837 void bpf_object__close(struct bpf_object *obj)
1838 {
1839         size_t i;
1840
1841         if (!obj)
1842                 return;
1843
1844         if (obj->clear_priv)
1845                 obj->clear_priv(obj, obj->priv);
1846
1847         bpf_object__elf_finish(obj);
1848         bpf_object__unload(obj);
1849         btf__free(obj->btf);
1850
1851         for (i = 0; i < obj->nr_maps; i++) {
1852                 zfree(&obj->maps[i].name);
1853                 if (obj->maps[i].clear_priv)
1854                         obj->maps[i].clear_priv(&obj->maps[i],
1855                                                 obj->maps[i].priv);
1856                 obj->maps[i].priv = NULL;
1857                 obj->maps[i].clear_priv = NULL;
1858         }
1859         zfree(&obj->maps);
1860         obj->nr_maps = 0;
1861
1862         if (obj->programs && obj->nr_programs) {
1863                 for (i = 0; i < obj->nr_programs; i++)
1864                         bpf_program__exit(&obj->programs[i]);
1865         }
1866         zfree(&obj->programs);
1867
1868         list_del(&obj->list);
1869         free(obj);
1870 }
1871
1872 struct bpf_object *
1873 bpf_object__next(struct bpf_object *prev)
1874 {
1875         struct bpf_object *next;
1876
1877         if (!prev)
1878                 next = list_first_entry(&bpf_objects_list,
1879                                         struct bpf_object,
1880                                         list);
1881         else
1882                 next = list_next_entry(prev, list);
1883
1884         /* Empty list is noticed here so don't need checking on entry. */
1885         if (&next->list == &bpf_objects_list)
1886                 return NULL;
1887
1888         return next;
1889 }
1890
1891 const char *bpf_object__name(struct bpf_object *obj)
1892 {
1893         return obj ? obj->path : ERR_PTR(-EINVAL);
1894 }
1895
1896 unsigned int bpf_object__kversion(struct bpf_object *obj)
1897 {
1898         return obj ? obj->kern_version : 0;
1899 }
1900
1901 int bpf_object__btf_fd(const struct bpf_object *obj)
1902 {
1903         return obj->btf ? btf__fd(obj->btf) : -1;
1904 }
1905
1906 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
1907                          bpf_object_clear_priv_t clear_priv)
1908 {
1909         if (obj->priv && obj->clear_priv)
1910                 obj->clear_priv(obj, obj->priv);
1911
1912         obj->priv = priv;
1913         obj->clear_priv = clear_priv;
1914         return 0;
1915 }
1916
1917 void *bpf_object__priv(struct bpf_object *obj)
1918 {
1919         return obj ? obj->priv : ERR_PTR(-EINVAL);
1920 }
1921
1922 static struct bpf_program *
1923 __bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1924 {
1925         size_t idx;
1926
1927         if (!obj->programs)
1928                 return NULL;
1929         /* First handler */
1930         if (prev == NULL)
1931                 return &obj->programs[0];
1932
1933         if (prev->obj != obj) {
1934                 pr_warning("error: program handler doesn't match object\n");
1935                 return NULL;
1936         }
1937
1938         idx = (prev - obj->programs) + 1;
1939         if (idx >= obj->nr_programs)
1940                 return NULL;
1941         return &obj->programs[idx];
1942 }
1943
1944 struct bpf_program *
1945 bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1946 {
1947         struct bpf_program *prog = prev;
1948
1949         do {
1950                 prog = __bpf_program__next(prog, obj);
1951         } while (prog && bpf_program__is_function_storage(prog, obj));
1952
1953         return prog;
1954 }
1955
1956 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
1957                           bpf_program_clear_priv_t clear_priv)
1958 {
1959         if (prog->priv && prog->clear_priv)
1960                 prog->clear_priv(prog, prog->priv);
1961
1962         prog->priv = priv;
1963         prog->clear_priv = clear_priv;
1964         return 0;
1965 }
1966
1967 void *bpf_program__priv(struct bpf_program *prog)
1968 {
1969         return prog ? prog->priv : ERR_PTR(-EINVAL);
1970 }
1971
1972 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
1973 {
1974         prog->prog_ifindex = ifindex;
1975 }
1976
1977 const char *bpf_program__title(struct bpf_program *prog, bool needs_copy)
1978 {
1979         const char *title;
1980
1981         title = prog->section_name;
1982         if (needs_copy) {
1983                 title = strdup(title);
1984                 if (!title) {
1985                         pr_warning("failed to strdup program title\n");
1986                         return ERR_PTR(-ENOMEM);
1987                 }
1988         }
1989
1990         return title;
1991 }
1992
1993 int bpf_program__fd(struct bpf_program *prog)
1994 {
1995         return bpf_program__nth_fd(prog, 0);
1996 }
1997
1998 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
1999                           bpf_program_prep_t prep)
2000 {
2001         int *instances_fds;
2002
2003         if (nr_instances <= 0 || !prep)
2004                 return -EINVAL;
2005
2006         if (prog->instances.nr > 0 || prog->instances.fds) {
2007                 pr_warning("Can't set pre-processor after loading\n");
2008                 return -EINVAL;
2009         }
2010
2011         instances_fds = malloc(sizeof(int) * nr_instances);
2012         if (!instances_fds) {
2013                 pr_warning("alloc memory failed for fds\n");
2014                 return -ENOMEM;
2015         }
2016
2017         /* fill all fd with -1 */
2018         memset(instances_fds, -1, sizeof(int) * nr_instances);
2019
2020         prog->instances.nr = nr_instances;
2021         prog->instances.fds = instances_fds;
2022         prog->preprocessor = prep;
2023         return 0;
2024 }
2025
2026 int bpf_program__nth_fd(struct bpf_program *prog, int n)
2027 {
2028         int fd;
2029
2030         if (!prog)
2031                 return -EINVAL;
2032
2033         if (n >= prog->instances.nr || n < 0) {
2034                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
2035                            n, prog->section_name, prog->instances.nr);
2036                 return -EINVAL;
2037         }
2038
2039         fd = prog->instances.fds[n];
2040         if (fd < 0) {
2041                 pr_warning("%dth instance of program '%s' is invalid\n",
2042                            n, prog->section_name);
2043                 return -ENOENT;
2044         }
2045
2046         return fd;
2047 }
2048
2049 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
2050 {
2051         prog->type = type;
2052 }
2053
2054 static bool bpf_program__is_type(struct bpf_program *prog,
2055                                  enum bpf_prog_type type)
2056 {
2057         return prog ? (prog->type == type) : false;
2058 }
2059
2060 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                   \
2061 int bpf_program__set_##NAME(struct bpf_program *prog)   \
2062 {                                                       \
2063         if (!prog)                                      \
2064                 return -EINVAL;                         \
2065         bpf_program__set_type(prog, TYPE);              \
2066         return 0;                                       \
2067 }                                                       \
2068                                                         \
2069 bool bpf_program__is_##NAME(struct bpf_program *prog)   \
2070 {                                                       \
2071         return bpf_program__is_type(prog, TYPE);        \
2072 }                                                       \
2073
2074 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
2075 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
2076 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
2077 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
2078 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
2079 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
2080 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
2081 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
2082
2083 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
2084                                            enum bpf_attach_type type)
2085 {
2086         prog->expected_attach_type = type;
2087 }
2088
2089 #define BPF_PROG_SEC_FULL(string, ptype, atype) \
2090         { string, sizeof(string) - 1, ptype, atype }
2091
2092 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_FULL(string, ptype, 0)
2093
2094 #define BPF_S_PROG_SEC(string, ptype) \
2095         BPF_PROG_SEC_FULL(string, BPF_PROG_TYPE_CGROUP_SOCK, ptype)
2096
2097 #define BPF_SA_PROG_SEC(string, ptype) \
2098         BPF_PROG_SEC_FULL(string, BPF_PROG_TYPE_CGROUP_SOCK_ADDR, ptype)
2099
2100 static const struct {
2101         const char *sec;
2102         size_t len;
2103         enum bpf_prog_type prog_type;
2104         enum bpf_attach_type expected_attach_type;
2105 } section_names[] = {
2106         BPF_PROG_SEC("socket",          BPF_PROG_TYPE_SOCKET_FILTER),
2107         BPF_PROG_SEC("kprobe/",         BPF_PROG_TYPE_KPROBE),
2108         BPF_PROG_SEC("kretprobe/",      BPF_PROG_TYPE_KPROBE),
2109         BPF_PROG_SEC("classifier",      BPF_PROG_TYPE_SCHED_CLS),
2110         BPF_PROG_SEC("action",          BPF_PROG_TYPE_SCHED_ACT),
2111         BPF_PROG_SEC("tracepoint/",     BPF_PROG_TYPE_TRACEPOINT),
2112         BPF_PROG_SEC("raw_tracepoint/", BPF_PROG_TYPE_RAW_TRACEPOINT),
2113         BPF_PROG_SEC("xdp",             BPF_PROG_TYPE_XDP),
2114         BPF_PROG_SEC("perf_event",      BPF_PROG_TYPE_PERF_EVENT),
2115         BPF_PROG_SEC("cgroup/skb",      BPF_PROG_TYPE_CGROUP_SKB),
2116         BPF_PROG_SEC("cgroup/sock",     BPF_PROG_TYPE_CGROUP_SOCK),
2117         BPF_PROG_SEC("cgroup/dev",      BPF_PROG_TYPE_CGROUP_DEVICE),
2118         BPF_PROG_SEC("lwt_in",          BPF_PROG_TYPE_LWT_IN),
2119         BPF_PROG_SEC("lwt_out",         BPF_PROG_TYPE_LWT_OUT),
2120         BPF_PROG_SEC("lwt_xmit",        BPF_PROG_TYPE_LWT_XMIT),
2121         BPF_PROG_SEC("lwt_seg6local",   BPF_PROG_TYPE_LWT_SEG6LOCAL),
2122         BPF_PROG_SEC("sockops",         BPF_PROG_TYPE_SOCK_OPS),
2123         BPF_PROG_SEC("sk_skb",          BPF_PROG_TYPE_SK_SKB),
2124         BPF_PROG_SEC("sk_msg",          BPF_PROG_TYPE_SK_MSG),
2125         BPF_PROG_SEC("lirc_mode2",      BPF_PROG_TYPE_LIRC_MODE2),
2126         BPF_SA_PROG_SEC("cgroup/bind4", BPF_CGROUP_INET4_BIND),
2127         BPF_SA_PROG_SEC("cgroup/bind6", BPF_CGROUP_INET6_BIND),
2128         BPF_SA_PROG_SEC("cgroup/connect4", BPF_CGROUP_INET4_CONNECT),
2129         BPF_SA_PROG_SEC("cgroup/connect6", BPF_CGROUP_INET6_CONNECT),
2130         BPF_SA_PROG_SEC("cgroup/sendmsg4", BPF_CGROUP_UDP4_SENDMSG),
2131         BPF_SA_PROG_SEC("cgroup/sendmsg6", BPF_CGROUP_UDP6_SENDMSG),
2132         BPF_S_PROG_SEC("cgroup/post_bind4", BPF_CGROUP_INET4_POST_BIND),
2133         BPF_S_PROG_SEC("cgroup/post_bind6", BPF_CGROUP_INET6_POST_BIND),
2134 };
2135
2136 #undef BPF_PROG_SEC
2137 #undef BPF_PROG_SEC_FULL
2138 #undef BPF_S_PROG_SEC
2139 #undef BPF_SA_PROG_SEC
2140
2141 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
2142                              enum bpf_attach_type *expected_attach_type)
2143 {
2144         int i;
2145
2146         if (!name)
2147                 return -EINVAL;
2148
2149         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
2150                 if (strncmp(name, section_names[i].sec, section_names[i].len))
2151                         continue;
2152                 *prog_type = section_names[i].prog_type;
2153                 *expected_attach_type = section_names[i].expected_attach_type;
2154                 return 0;
2155         }
2156         return -EINVAL;
2157 }
2158
2159 static int
2160 bpf_program__identify_section(struct bpf_program *prog,
2161                               enum bpf_prog_type *prog_type,
2162                               enum bpf_attach_type *expected_attach_type)
2163 {
2164         return libbpf_prog_type_by_name(prog->section_name, prog_type,
2165                                         expected_attach_type);
2166 }
2167
2168 int bpf_map__fd(struct bpf_map *map)
2169 {
2170         return map ? map->fd : -EINVAL;
2171 }
2172
2173 const struct bpf_map_def *bpf_map__def(struct bpf_map *map)
2174 {
2175         return map ? &map->def : ERR_PTR(-EINVAL);
2176 }
2177
2178 const char *bpf_map__name(struct bpf_map *map)
2179 {
2180         return map ? map->name : NULL;
2181 }
2182
2183 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
2184 {
2185         return map ? map->btf_key_type_id : 0;
2186 }
2187
2188 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
2189 {
2190         return map ? map->btf_value_type_id : 0;
2191 }
2192
2193 int bpf_map__set_priv(struct bpf_map *map, void *priv,
2194                      bpf_map_clear_priv_t clear_priv)
2195 {
2196         if (!map)
2197                 return -EINVAL;
2198
2199         if (map->priv) {
2200                 if (map->clear_priv)
2201                         map->clear_priv(map, map->priv);
2202         }
2203
2204         map->priv = priv;
2205         map->clear_priv = clear_priv;
2206         return 0;
2207 }
2208
2209 void *bpf_map__priv(struct bpf_map *map)
2210 {
2211         return map ? map->priv : ERR_PTR(-EINVAL);
2212 }
2213
2214 bool bpf_map__is_offload_neutral(struct bpf_map *map)
2215 {
2216         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
2217 }
2218
2219 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
2220 {
2221         map->map_ifindex = ifindex;
2222 }
2223
2224 struct bpf_map *
2225 bpf_map__next(struct bpf_map *prev, struct bpf_object *obj)
2226 {
2227         size_t idx;
2228         struct bpf_map *s, *e;
2229
2230         if (!obj || !obj->maps)
2231                 return NULL;
2232
2233         s = obj->maps;
2234         e = obj->maps + obj->nr_maps;
2235
2236         if (prev == NULL)
2237                 return s;
2238
2239         if ((prev < s) || (prev >= e)) {
2240                 pr_warning("error in %s: map handler doesn't belong to object\n",
2241                            __func__);
2242                 return NULL;
2243         }
2244
2245         idx = (prev - obj->maps) + 1;
2246         if (idx >= obj->nr_maps)
2247                 return NULL;
2248         return &obj->maps[idx];
2249 }
2250
2251 struct bpf_map *
2252 bpf_object__find_map_by_name(struct bpf_object *obj, const char *name)
2253 {
2254         struct bpf_map *pos;
2255
2256         bpf_map__for_each(pos, obj) {
2257                 if (pos->name && !strcmp(pos->name, name))
2258                         return pos;
2259         }
2260         return NULL;
2261 }
2262
2263 struct bpf_map *
2264 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
2265 {
2266         int i;
2267
2268         for (i = 0; i < obj->nr_maps; i++) {
2269                 if (obj->maps[i].offset == offset)
2270                         return &obj->maps[i];
2271         }
2272         return ERR_PTR(-ENOENT);
2273 }
2274
2275 long libbpf_get_error(const void *ptr)
2276 {
2277         if (IS_ERR(ptr))
2278                 return PTR_ERR(ptr);
2279         return 0;
2280 }
2281
2282 int bpf_prog_load(const char *file, enum bpf_prog_type type,
2283                   struct bpf_object **pobj, int *prog_fd)
2284 {
2285         struct bpf_prog_load_attr attr;
2286
2287         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
2288         attr.file = file;
2289         attr.prog_type = type;
2290         attr.expected_attach_type = 0;
2291
2292         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
2293 }
2294
2295 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
2296                         struct bpf_object **pobj, int *prog_fd)
2297 {
2298         struct bpf_object_open_attr open_attr = {};
2299         struct bpf_program *prog, *first_prog = NULL;
2300         enum bpf_attach_type expected_attach_type;
2301         enum bpf_prog_type prog_type;
2302         struct bpf_object *obj;
2303         struct bpf_map *map;
2304         int err;
2305
2306         if (!attr)
2307                 return -EINVAL;
2308         if (!attr->file)
2309                 return -EINVAL;
2310
2311         open_attr.file = attr->file;
2312         open_attr.prog_type = attr->prog_type;
2313
2314         obj = bpf_object__open_xattr(&open_attr);
2315         if (IS_ERR_OR_NULL(obj))
2316                 return -ENOENT;
2317
2318         bpf_object__for_each_program(prog, obj) {
2319                 /*
2320                  * If type is not specified, try to guess it based on
2321                  * section name.
2322                  */
2323                 prog_type = attr->prog_type;
2324                 prog->prog_ifindex = attr->ifindex;
2325                 expected_attach_type = attr->expected_attach_type;
2326                 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
2327                         err = bpf_program__identify_section(prog, &prog_type,
2328                                                             &expected_attach_type);
2329                         if (err < 0) {
2330                                 pr_warning("failed to guess program type based on section name %s\n",
2331                                            prog->section_name);
2332                                 bpf_object__close(obj);
2333                                 return -EINVAL;
2334                         }
2335                 }
2336
2337                 bpf_program__set_type(prog, prog_type);
2338                 bpf_program__set_expected_attach_type(prog,
2339                                                       expected_attach_type);
2340
2341                 if (!bpf_program__is_function_storage(prog, obj) && !first_prog)
2342                         first_prog = prog;
2343         }
2344
2345         bpf_map__for_each(map, obj) {
2346                 if (!bpf_map__is_offload_neutral(map))
2347                         map->map_ifindex = attr->ifindex;
2348         }
2349
2350         if (!first_prog) {
2351                 pr_warning("object file doesn't contain bpf program\n");
2352                 bpf_object__close(obj);
2353                 return -ENOENT;
2354         }
2355
2356         err = bpf_object__load(obj);
2357         if (err) {
2358                 bpf_object__close(obj);
2359                 return -EINVAL;
2360         }
2361
2362         *pobj = obj;
2363         *prog_fd = bpf_program__fd(first_prog);
2364         return 0;
2365 }
2366
2367 enum bpf_perf_event_ret
2368 bpf_perf_event_read_simple(void *mem, unsigned long size,
2369                            unsigned long page_size, void **buf, size_t *buf_len,
2370                            bpf_perf_event_print_t fn, void *priv)
2371 {
2372         volatile struct perf_event_mmap_page *header = mem;
2373         __u64 data_tail = header->data_tail;
2374         __u64 data_head = header->data_head;
2375         int ret = LIBBPF_PERF_EVENT_ERROR;
2376         void *base, *begin, *end;
2377
2378         asm volatile("" ::: "memory"); /* in real code it should be smp_rmb() */
2379         if (data_head == data_tail)
2380                 return LIBBPF_PERF_EVENT_CONT;
2381
2382         base = ((char *)header) + page_size;
2383
2384         begin = base + data_tail % size;
2385         end = base + data_head % size;
2386
2387         while (begin != end) {
2388                 struct perf_event_header *ehdr;
2389
2390                 ehdr = begin;
2391                 if (begin + ehdr->size > base + size) {
2392                         long len = base + size - begin;
2393
2394                         if (*buf_len < ehdr->size) {
2395                                 free(*buf);
2396                                 *buf = malloc(ehdr->size);
2397                                 if (!*buf) {
2398                                         ret = LIBBPF_PERF_EVENT_ERROR;
2399                                         break;
2400                                 }
2401                                 *buf_len = ehdr->size;
2402                         }
2403
2404                         memcpy(*buf, begin, len);
2405                         memcpy(*buf + len, base, ehdr->size - len);
2406                         ehdr = (void *)*buf;
2407                         begin = base + ehdr->size - len;
2408                 } else if (begin + ehdr->size == base + size) {
2409                         begin = base;
2410                 } else {
2411                         begin += ehdr->size;
2412                 }
2413
2414                 ret = fn(ehdr, priv);
2415                 if (ret != LIBBPF_PERF_EVENT_CONT)
2416                         break;
2417
2418                 data_tail += ehdr->size;
2419         }
2420
2421         __sync_synchronize(); /* smp_mb() */
2422         header->data_tail = data_tail;
2423
2424         return ret;
2425 }