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 #define _GNU_SOURCE
  26 #include <stdlib.h>
  27 #include <stdio.h>
  28 #include <stdarg.h>
  29 #include <libgen.h>
  30 #include <inttypes.h>
  31 #include <string.h>
  32 #include <unistd.h>
  33 #include <fcntl.h>
  34 #include <errno.h>
  35 #include <perf-sys.h>
  36 #include <asm/unistd.h>
  37 #include <linux/err.h>
  38 #include <linux/kernel.h>
  39 #include <linux/bpf.h>
  40 #include <linux/btf.h>
  41 #include <linux/list.h>
  42 #include <linux/limits.h>
  43 #include <sys/stat.h>
  44 #include <sys/types.h>
  45 #include <sys/vfs.h>
  46 #include <tools/libc_compat.h>
  47 #include <libelf.h>
  48 #include <gelf.h>
  49
  50 #include "libbpf.h"
  51 #include "bpf.h"
  52 #include "btf.h"
  53 #include "str_error.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 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  64
  65 __printf(1, 2)
  66 static int __base_pr(const char *format, ...)
  67 {
  68         va_list args;
  69         int err;
  70
  71         va_start(args, format);
  72         err = vfprintf(stderr, format, args);
  73         va_end(args);
  74         return err;
  75 }
  76
  77 static __printf(1, 2) libbpf_print_fn_t __pr_warning = __base_pr;
  78 static __printf(1, 2) libbpf_print_fn_t __pr_info = __base_pr;
  79 static __printf(1, 2) libbpf_print_fn_t __pr_debug;
  80
  81 #define __pr(func, fmt, ...)    \
  82 do {                            \
  83         if ((func))             \
  84                 (func)("libbpf: " fmt, ##__VA_ARGS__); \
  85 } while (0)
  86
  87 #define pr_warning(fmt, ...)    __pr(__pr_warning, fmt, ##__VA_ARGS__)
  88 #define pr_info(fmt, ...)       __pr(__pr_info, fmt, ##__VA_ARGS__)
  89 #define pr_debug(fmt, ...)      __pr(__pr_debug, fmt, ##__VA_ARGS__)
  90
  91 void libbpf_set_print(libbpf_print_fn_t warn,
  92                       libbpf_print_fn_t info,
  93                       libbpf_print_fn_t debug)
  94 {
  95         __pr_warning = warn;
  96         __pr_info = info;
  97         __pr_debug = debug;
  98 }
  99
 100 #define STRERR_BUFSIZE  128
 101
 102 #define CHECK_ERR(action, err, out) do {        \
 103         err = action;                   \
 104         if (err)                        \
 105                 goto out;               \
 106 } while(0)
 107
 108
 109 /* Copied from tools/perf/util/util.h */
 110 #ifndef zfree
 111 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 112 #endif
 113
 114 #ifndef zclose
 115 # define zclose(fd) ({                  \
 116         int ___err = 0;                 \
 117         if ((fd) >= 0)                  \
 118                 ___err = close((fd));   \
 119         fd = -1;                        \
 120         ___err; })
 121 #endif
 122
 123 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 124 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 125 #else
 126 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 127 #endif
 128
 129 /*
 130  * bpf_prog should be a better name but it has been used in
 131  * linux/filter.h.
 132  */
 133 struct bpf_program {
 134         /* Index in elf obj file, for relocation use. */
 135         int idx;
 136         char *name;
 137         int prog_ifindex;
 138         char *section_name;
 139         struct bpf_insn *insns;
 140         size_t insns_cnt, main_prog_cnt;
 141         enum bpf_prog_type type;
 142
 143         struct reloc_desc {
 144                 enum {
 145                         RELO_LD64,
 146                         RELO_CALL,
 147                 } type;
 148                 int insn_idx;
 149                 union {
 150                         int map_idx;
 151                         int text_off;
 152                 };
 153         } *reloc_desc;
 154         int nr_reloc;
 155
 156         struct {
 157                 int nr;
 158                 int *fds;
 159         } instances;
 160         bpf_program_prep_t preprocessor;
 161
 162         struct bpf_object *obj;
 163         void *priv;
 164         bpf_program_clear_priv_t clear_priv;
 165
 166         enum bpf_attach_type expected_attach_type;
 167 };
 168
 169 struct bpf_map {
 170         int fd;
 171         char *name;
 172         size_t offset;
 173         int map_ifindex;
 174         struct bpf_map_def def;
 175         __u32 btf_key_type_id;
 176         __u32 btf_value_type_id;
 177         void *priv;
 178         bpf_map_clear_priv_t clear_priv;
 179 };
 180
 181 static LIST_HEAD(bpf_objects_list);
 182
 183 struct bpf_object {
 184         char license[64];
 185         u32 kern_version;
 186
 187         struct bpf_program *programs;
 188         size_t nr_programs;
 189         struct bpf_map *maps;
 190         size_t nr_maps;
 191
 192         bool loaded;
 193         bool has_pseudo_calls;
 194
 195         /*
 196          * Information when doing elf related work. Only valid if fd
 197          * is valid.
 198          */
 199         struct {
 200                 int fd;
 201                 void *obj_buf;
 202                 size_t obj_buf_sz;
 203                 Elf *elf;
 204                 GElf_Ehdr ehdr;
 205                 Elf_Data *symbols;
 206                 size_t strtabidx;
 207                 struct {
 208                         GElf_Shdr shdr;
 209                         Elf_Data *data;
 210                 } *reloc;
 211                 int nr_reloc;
 212                 int maps_shndx;
 213                 int text_shndx;
 214         } efile;
 215         /*
 216          * All loaded bpf_object is linked in a list, which is
 217          * hidden to caller. bpf_objects__<func> handlers deal with
 218          * all objects.
 219          */
 220         struct list_head list;
 221
 222         struct btf *btf;
 223
 224         void *priv;
 225         bpf_object_clear_priv_t clear_priv;
 226
 227         char path[];
 228 };
 229 #define obj_elf_valid(o)        ((o)->efile.elf)
 230
 231 void bpf_program__unload(struct bpf_program *prog)
 232 {
 233         int i;
 234
 235         if (!prog)
 236                 return;
 237
 238         /*
 239          * If the object is opened but the program was never loaded,
 240          * it is possible that prog->instances.nr == -1.
 241          */
 242         if (prog->instances.nr > 0) {
 243                 for (i = 0; i < prog->instances.nr; i++)
 244                         zclose(prog->instances.fds[i]);
 245         } else if (prog->instances.nr != -1) {
 246                 pr_warning("Internal error: instances.nr is %d\n",
 247                            prog->instances.nr);
 248         }
 249
 250         prog->instances.nr = -1;
 251         zfree(&prog->instances.fds);
 252 }
 253
 254 static void bpf_program__exit(struct bpf_program *prog)
 255 {
 256         if (!prog)
 257                 return;
 258
 259         if (prog->clear_priv)
 260                 prog->clear_priv(prog, prog->priv);
 261
 262         prog->priv = NULL;
 263         prog->clear_priv = NULL;
 264
 265         bpf_program__unload(prog);
 266         zfree(&prog->name);
 267         zfree(&prog->section_name);
 268         zfree(&prog->insns);
 269         zfree(&prog->reloc_desc);
 270
 271         prog->nr_reloc = 0;
 272         prog->insns_cnt = 0;
 273         prog->idx = -1;
 274 }
 275
 276 static int
 277 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 278                   struct bpf_program *prog)
 279 {
 280         if (size < sizeof(struct bpf_insn)) {
 281                 pr_warning("corrupted section '%s'\n", section_name);
 282                 return -EINVAL;
 283         }
 284
 285         bzero(prog, sizeof(*prog));
 286
 287         prog->section_name = strdup(section_name);
 288         if (!prog->section_name) {
 289                 pr_warning("failed to alloc name for prog under section(%d) %s\n",
 290                            idx, section_name);
 291                 goto errout;
 292         }
 293
 294         prog->insns = malloc(size);
 295         if (!prog->insns) {
 296                 pr_warning("failed to alloc insns for prog under section %s\n",
 297                            section_name);
 298                 goto errout;
 299         }
 300         prog->insns_cnt = size / sizeof(struct bpf_insn);
 301         memcpy(prog->insns, data,
 302                prog->insns_cnt * sizeof(struct bpf_insn));
 303         prog->idx = idx;
 304         prog->instances.fds = NULL;
 305         prog->instances.nr = -1;
 306         prog->type = BPF_PROG_TYPE_KPROBE;
 307
 308         return 0;
 309 errout:
 310         bpf_program__exit(prog);
 311         return -ENOMEM;
 312 }
 313
 314 static int
 315 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 316                         char *section_name, int idx)
 317 {
 318         struct bpf_program prog, *progs;
 319         int nr_progs, err;
 320
 321         err = bpf_program__init(data, size, section_name, idx, &prog);
 322         if (err)
 323                 return err;
 324
 325         progs = obj->programs;
 326         nr_progs = obj->nr_programs;
 327
 328         progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
 329         if (!progs) {
 330                 /*
 331                  * In this case the original obj->programs
 332                  * is still valid, so don't need special treat for
 333                  * bpf_close_object().
 334                  */
 335                 pr_warning("failed to alloc a new program under section '%s'\n",
 336                            section_name);
 337                 bpf_program__exit(&prog);
 338                 return -ENOMEM;
 339         }
 340
 341         pr_debug("found program %s\n", prog.section_name);
 342         obj->programs = progs;
 343         obj->nr_programs = nr_progs + 1;
 344         prog.obj = obj;
 345         progs[nr_progs] = prog;
 346         return 0;
 347 }
 348
 349 static int
 350 bpf_object__init_prog_names(struct bpf_object *obj)
 351 {
 352         Elf_Data *symbols = obj->efile.symbols;
 353         struct bpf_program *prog;
 354         size_t pi, si;
 355
 356         for (pi = 0; pi < obj->nr_programs; pi++) {
 357                 const char *name = NULL;
 358
 359                 prog = &obj->programs[pi];
 360
 361                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 362                      si++) {
 363                         GElf_Sym sym;
 364
 365                         if (!gelf_getsym(symbols, si, &sym))
 366                                 continue;
 367                         if (sym.st_shndx != prog->idx)
 368                                 continue;
 369                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 370                                 continue;
 371
 372                         name = elf_strptr(obj->efile.elf,
 373                                           obj->efile.strtabidx,
 374                                           sym.st_name);
 375                         if (!name) {
 376                                 pr_warning("failed to get sym name string for prog %s\n",
 377                                            prog->section_name);
 378                                 return -LIBBPF_ERRNO__LIBELF;
 379                         }
 380                 }
 381
 382                 if (!name && prog->idx == obj->efile.text_shndx)
 383                         name = ".text";
 384
 385                 if (!name) {
 386                         pr_warning("failed to find sym for prog %s\n",
 387                                    prog->section_name);
 388                         return -EINVAL;
 389                 }
 390
 391                 prog->name = strdup(name);
 392                 if (!prog->name) {
 393                         pr_warning("failed to allocate memory for prog sym %s\n",
 394                                    name);
 395                         return -ENOMEM;
 396                 }
 397         }
 398
 399         return 0;
 400 }
 401
 402 static struct bpf_object *bpf_object__new(const char *path,
 403                                           void *obj_buf,
 404                                           size_t obj_buf_sz)
 405 {
 406         struct bpf_object *obj;
 407
 408         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 409         if (!obj) {
 410                 pr_warning("alloc memory failed for %s\n", path);
 411                 return ERR_PTR(-ENOMEM);
 412         }
 413
 414         strcpy(obj->path, path);
 415         obj->efile.fd = -1;
 416
 417         /*
 418          * Caller of this function should also calls
 419          * bpf_object__elf_finish() after data collection to return
 420          * obj_buf to user. If not, we should duplicate the buffer to
 421          * avoid user freeing them before elf finish.
 422          */
 423         obj->efile.obj_buf = obj_buf;
 424         obj->efile.obj_buf_sz = obj_buf_sz;
 425         obj->efile.maps_shndx = -1;
 426
 427         obj->loaded = false;
 428
 429         INIT_LIST_HEAD(&obj->list);
 430         list_add(&obj->list, &bpf_objects_list);
 431         return obj;
 432 }
 433
 434 static void bpf_object__elf_finish(struct bpf_object *obj)
 435 {
 436         if (!obj_elf_valid(obj))
 437                 return;
 438
 439         if (obj->efile.elf) {
 440                 elf_end(obj->efile.elf);
 441                 obj->efile.elf = NULL;
 442         }
 443         obj->efile.symbols = NULL;
 444
 445         zfree(&obj->efile.reloc);
 446         obj->efile.nr_reloc = 0;
 447         zclose(obj->efile.fd);
 448         obj->efile.obj_buf = NULL;
 449         obj->efile.obj_buf_sz = 0;
 450 }
 451
 452 static int bpf_object__elf_init(struct bpf_object *obj)
 453 {
 454         int err = 0;
 455         GElf_Ehdr *ep;
 456
 457         if (obj_elf_valid(obj)) {
 458                 pr_warning("elf init: internal error\n");
 459                 return -LIBBPF_ERRNO__LIBELF;
 460         }
 461
 462         if (obj->efile.obj_buf_sz > 0) {
 463                 /*
 464                  * obj_buf should have been validated by
 465                  * bpf_object__open_buffer().
 466                  */
 467                 obj->efile.elf = elf_memory(obj->efile.obj_buf,
 468                                             obj->efile.obj_buf_sz);
 469         } else {
 470                 obj->efile.fd = open(obj->path, O_RDONLY);
 471                 if (obj->efile.fd < 0) {
 472                         char errmsg[STRERR_BUFSIZE];
 473                         char *cp = libbpf_strerror_r(errno, errmsg,
 474                                                      sizeof(errmsg));
 475
 476                         pr_warning("failed to open %s: %s\n", obj->path, cp);
 477                         return -errno;
 478                 }
 479
 480                 obj->efile.elf = elf_begin(obj->efile.fd,
 481                                 LIBBPF_ELF_C_READ_MMAP,
 482                                 NULL);
 483         }
 484
 485         if (!obj->efile.elf) {
 486                 pr_warning("failed to open %s as ELF file\n",
 487                                 obj->path);
 488                 err = -LIBBPF_ERRNO__LIBELF;
 489                 goto errout;
 490         }
 491
 492         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 493                 pr_warning("failed to get EHDR from %s\n",
 494                                 obj->path);
 495                 err = -LIBBPF_ERRNO__FORMAT;
 496                 goto errout;
 497         }
 498         ep = &obj->efile.ehdr;
 499
 500         /* Old LLVM set e_machine to EM_NONE */
 501         if ((ep->e_type != ET_REL) || (ep->e_machine && (ep->e_machine != EM_BPF))) {
 502                 pr_warning("%s is not an eBPF object file\n",
 503                         obj->path);
 504                 err = -LIBBPF_ERRNO__FORMAT;
 505                 goto errout;
 506         }
 507
 508         return 0;
 509 errout:
 510         bpf_object__elf_finish(obj);
 511         return err;
 512 }
 513
 514 static int
 515 bpf_object__check_endianness(struct bpf_object *obj)
 516 {
 517         static unsigned int const endian = 1;
 518
 519         switch (obj->efile.ehdr.e_ident[EI_DATA]) {
 520         case ELFDATA2LSB:
 521                 /* We are big endian, BPF obj is little endian. */
 522                 if (*(unsigned char const *)&endian != 1)
 523                         goto mismatch;
 524                 break;
 525
 526         case ELFDATA2MSB:
 527                 /* We are little endian, BPF obj is big endian. */
 528                 if (*(unsigned char const *)&endian != 0)
 529                         goto mismatch;
 530                 break;
 531         default:
 532                 return -LIBBPF_ERRNO__ENDIAN;
 533         }
 534
 535         return 0;
 536
 537 mismatch:
 538         pr_warning("Error: endianness mismatch.\n");
 539         return -LIBBPF_ERRNO__ENDIAN;
 540 }
 541
 542 static int
 543 bpf_object__init_license(struct bpf_object *obj,
 544                          void *data, size_t size)
 545 {
 546         memcpy(obj->license, data,
 547                min(size, sizeof(obj->license) - 1));
 548         pr_debug("license of %s is %s\n", obj->path, obj->license);
 549         return 0;
 550 }
 551
 552 static int
 553 bpf_object__init_kversion(struct bpf_object *obj,
 554                           void *data, size_t size)
 555 {
 556         u32 kver;
 557
 558         if (size != sizeof(kver)) {
 559                 pr_warning("invalid kver section in %s\n", obj->path);
 560                 return -LIBBPF_ERRNO__FORMAT;
 561         }
 562         memcpy(&kver, data, sizeof(kver));
 563         obj->kern_version = kver;
 564         pr_debug("kernel version of %s is %x\n", obj->path,
 565                  obj->kern_version);
 566         return 0;
 567 }
 568
 569 static int compare_bpf_map(const void *_a, const void *_b)
 570 {
 571         const struct bpf_map *a = _a;
 572         const struct bpf_map *b = _b;
 573
 574         return a->offset - b->offset;
 575 }
 576
 577 static int
 578 bpf_object__init_maps(struct bpf_object *obj)
 579 {
 580         int i, map_idx, map_def_sz, nr_maps = 0;
 581         Elf_Scn *scn;
 582         Elf_Data *data;
 583         Elf_Data *symbols = obj->efile.symbols;
 584
 585         if (obj->efile.maps_shndx < 0)
 586                 return -EINVAL;
 587         if (!symbols)
 588                 return -EINVAL;
 589
 590         scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
 591         if (scn)
 592                 data = elf_getdata(scn, NULL);
 593         if (!scn || !data) {
 594                 pr_warning("failed to get Elf_Data from map section %d\n",
 595                            obj->efile.maps_shndx);
 596                 return -EINVAL;
 597         }
 598
 599         /*
 600          * Count number of maps. Each map has a name.
 601          * Array of maps is not supported: only the first element is
 602          * considered.
 603          *
 604          * TODO: Detect array of map and report error.
 605          */
 606         for (i = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 607                 GElf_Sym sym;
 608
 609                 if (!gelf_getsym(symbols, i, &sym))
 610                         continue;
 611                 if (sym.st_shndx != obj->efile.maps_shndx)
 612                         continue;
 613                 nr_maps++;
 614         }
 615
 616         /* Alloc obj->maps and fill nr_maps. */
 617         pr_debug("maps in %s: %d maps in %zd bytes\n", obj->path,
 618                  nr_maps, data->d_size);
 619
 620         if (!nr_maps)
 621                 return 0;
 622
 623         /* Assume equally sized map definitions */
 624         map_def_sz = data->d_size / nr_maps;
 625         if (!data->d_size || (data->d_size % nr_maps) != 0) {
 626                 pr_warning("unable to determine map definition size "
 627                            "section %s, %d maps in %zd bytes\n",
 628                            obj->path, nr_maps, data->d_size);
 629                 return -EINVAL;
 630         }
 631
 632         obj->maps = calloc(nr_maps, sizeof(obj->maps[0]));
 633         if (!obj->maps) {
 634                 pr_warning("alloc maps for object failed\n");
 635                 return -ENOMEM;
 636         }
 637         obj->nr_maps = nr_maps;
 638
 639         /*
 640          * fill all fd with -1 so won't close incorrect
 641          * fd (fd=0 is stdin) when failure (zclose won't close
 642          * negative fd)).
 643          */
 644         for (i = 0; i < nr_maps; i++)
 645                 obj->maps[i].fd = -1;
 646
 647         /*
 648          * Fill obj->maps using data in "maps" section.
 649          */
 650         for (i = 0, map_idx = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 651                 GElf_Sym sym;
 652                 const char *map_name;
 653                 struct bpf_map_def *def;
 654
 655                 if (!gelf_getsym(symbols, i, &sym))
 656                         continue;
 657                 if (sym.st_shndx != obj->efile.maps_shndx)
 658                         continue;
 659
 660                 map_name = elf_strptr(obj->efile.elf,
 661                                       obj->efile.strtabidx,
 662                                       sym.st_name);
 663                 obj->maps[map_idx].offset = sym.st_value;
 664                 if (sym.st_value + map_def_sz > data->d_size) {
 665                         pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
 666                                    obj->path, map_name);
 667                         return -EINVAL;
 668                 }
 669
 670                 obj->maps[map_idx].name = strdup(map_name);
 671                 if (!obj->maps[map_idx].name) {
 672                         pr_warning("failed to alloc map name\n");
 673                         return -ENOMEM;
 674                 }
 675                 pr_debug("map %d is \"%s\"\n", map_idx,
 676                          obj->maps[map_idx].name);
 677                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
 678                 /*
 679                  * If the definition of the map in the object file fits in
 680                  * bpf_map_def, copy it.  Any extra fields in our version
 681                  * of bpf_map_def will default to zero as a result of the
 682                  * calloc above.
 683                  */
 684                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
 685                         memcpy(&obj->maps[map_idx].def, def, map_def_sz);
 686                 } else {
 687                         /*
 688                          * Here the map structure being read is bigger than what
 689                          * we expect, truncate if the excess bits are all zero.
 690                          * If they are not zero, reject this map as
 691                          * incompatible.
 692                          */
 693                         char *b;
 694                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
 695                              b < ((char *)def) + map_def_sz; b++) {
 696                                 if (*b != 0) {
 697                                         pr_warning("maps section in %s: \"%s\" "
 698                                                    "has unrecognized, non-zero "
 699                                                    "options\n",
 700                                                    obj->path, map_name);
 701                                         return -EINVAL;
 702                                 }
 703                         }
 704                         memcpy(&obj->maps[map_idx].def, def,
 705                                sizeof(struct bpf_map_def));
 706                 }
 707                 map_idx++;
 708         }
 709
 710         qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]), compare_bpf_map);
 711         return 0;
 712 }
 713
 714 static bool section_have_execinstr(struct bpf_object *obj, int idx)
 715 {
 716         Elf_Scn *scn;
 717         GElf_Shdr sh;
 718
 719         scn = elf_getscn(obj->efile.elf, idx);
 720         if (!scn)
 721                 return false;
 722
 723         if (gelf_getshdr(scn, &sh) != &sh)
 724                 return false;
 725
 726         if (sh.sh_flags & SHF_EXECINSTR)
 727                 return true;
 728
 729         return false;
 730 }
 731
 732 static int bpf_object__elf_collect(struct bpf_object *obj)
 733 {
 734         Elf *elf = obj->efile.elf;
 735         GElf_Ehdr *ep = &obj->efile.ehdr;
 736         Elf_Scn *scn = NULL;
 737         int idx = 0, err = 0;
 738
 739         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
 740         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
 741                 pr_warning("failed to get e_shstrndx from %s\n",
 742                            obj->path);
 743                 return -LIBBPF_ERRNO__FORMAT;
 744         }
 745
 746         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 747                 char *name;
 748                 GElf_Shdr sh;
 749                 Elf_Data *data;
 750
 751                 idx++;
 752                 if (gelf_getshdr(scn, &sh) != &sh) {
 753                         pr_warning("failed to get section(%d) header from %s\n",
 754                                    idx, obj->path);
 755                         err = -LIBBPF_ERRNO__FORMAT;
 756                         goto out;
 757                 }
 758
 759                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 760                 if (!name) {
 761                         pr_warning("failed to get section(%d) name from %s\n",
 762                                    idx, obj->path);
 763                         err = -LIBBPF_ERRNO__FORMAT;
 764                         goto out;
 765                 }
 766
 767                 data = elf_getdata(scn, 0);
 768                 if (!data) {
 769                         pr_warning("failed to get section(%d) data from %s(%s)\n",
 770                                    idx, name, obj->path);
 771                         err = -LIBBPF_ERRNO__FORMAT;
 772                         goto out;
 773                 }
 774                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
 775                          idx, name, (unsigned long)data->d_size,
 776                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
 777                          (int)sh.sh_type);
 778
 779                 if (strcmp(name, "license") == 0)
 780                         err = bpf_object__init_license(obj,
 781                                                        data->d_buf,
 782                                                        data->d_size);
 783                 else if (strcmp(name, "version") == 0)
 784                         err = bpf_object__init_kversion(obj,
 785                                                         data->d_buf,
 786                                                         data->d_size);
 787                 else if (strcmp(name, "maps") == 0)
 788                         obj->efile.maps_shndx = idx;
 789                 else if (strcmp(name, BTF_ELF_SEC) == 0) {
 790                         obj->btf = btf__new(data->d_buf, data->d_size,
 791                                             __pr_debug);
 792                         if (IS_ERR(obj->btf)) {
 793                                 pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
 794                                            BTF_ELF_SEC, PTR_ERR(obj->btf));
 795                                 obj->btf = NULL;
 796                         }
 797                 } else if (sh.sh_type == SHT_SYMTAB) {
 798                         if (obj->efile.symbols) {
 799                                 pr_warning("bpf: multiple SYMTAB in %s\n",
 800                                            obj->path);
 801                                 err = -LIBBPF_ERRNO__FORMAT;
 802                         } else {
 803                                 obj->efile.symbols = data;
 804                                 obj->efile.strtabidx = sh.sh_link;
 805                         }
 806                 } else if ((sh.sh_type == SHT_PROGBITS) &&
 807                            (sh.sh_flags & SHF_EXECINSTR) &&
 808                            (data->d_size > 0)) {
 809                         if (strcmp(name, ".text") == 0)
 810                                 obj->efile.text_shndx = idx;
 811                         err = bpf_object__add_program(obj, data->d_buf,
 812                                                       data->d_size, name, idx);
 813                         if (err) {
 814                                 char errmsg[STRERR_BUFSIZE];
 815                                 char *cp = libbpf_strerror_r(-err, errmsg,
 816                                                              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 = libbpf_strerror_r(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 = libbpf_strerror_r(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 = libbpf_strerror_r(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 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         case BPF_PROG_TYPE_FLOW_DISSECTOR:
1508                 return false;
1509         case BPF_PROG_TYPE_UNSPEC:
1510         case BPF_PROG_TYPE_KPROBE:
1511         case BPF_PROG_TYPE_TRACEPOINT:
1512         case BPF_PROG_TYPE_PERF_EVENT:
1513         case BPF_PROG_TYPE_RAW_TRACEPOINT:
1514         default:
1515                 return true;
1516         }
1517 }
1518
1519 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
1520 {
1521         if (needs_kver && obj->kern_version == 0) {
1522                 pr_warning("%s doesn't provide kernel version\n",
1523                            obj->path);
1524                 return -LIBBPF_ERRNO__KVERSION;
1525         }
1526         return 0;
1527 }
1528
1529 static struct bpf_object *
1530 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
1531                    bool needs_kver)
1532 {
1533         struct bpf_object *obj;
1534         int err;
1535
1536         if (elf_version(EV_CURRENT) == EV_NONE) {
1537                 pr_warning("failed to init libelf for %s\n", path);
1538                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1539         }
1540
1541         obj = bpf_object__new(path, obj_buf, obj_buf_sz);
1542         if (IS_ERR(obj))
1543                 return obj;
1544
1545         CHECK_ERR(bpf_object__elf_init(obj), err, out);
1546         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
1547         CHECK_ERR(bpf_object__elf_collect(obj), err, out);
1548         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
1549         CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
1550
1551         bpf_object__elf_finish(obj);
1552         return obj;
1553 out:
1554         bpf_object__close(obj);
1555         return ERR_PTR(err);
1556 }
1557
1558 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
1559 {
1560         /* param validation */
1561         if (!attr->file)
1562                 return NULL;
1563
1564         pr_debug("loading %s\n", attr->file);
1565
1566         return __bpf_object__open(attr->file, NULL, 0,
1567                                   bpf_prog_type__needs_kver(attr->prog_type));
1568 }
1569
1570 struct bpf_object *bpf_object__open(const char *path)
1571 {
1572         struct bpf_object_open_attr attr = {
1573                 .file           = path,
1574                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
1575         };
1576
1577         return bpf_object__open_xattr(&attr);
1578 }
1579
1580 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
1581                                            size_t obj_buf_sz,
1582                                            const char *name)
1583 {
1584         char tmp_name[64];
1585
1586         /* param validation */
1587         if (!obj_buf || obj_buf_sz <= 0)
1588                 return NULL;
1589
1590         if (!name) {
1591                 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
1592                          (unsigned long)obj_buf,
1593                          (unsigned long)obj_buf_sz);
1594                 tmp_name[sizeof(tmp_name) - 1] = '\0';
1595                 name = tmp_name;
1596         }
1597         pr_debug("loading object '%s' from buffer\n",
1598                  name);
1599
1600         return __bpf_object__open(name, obj_buf, obj_buf_sz, true);
1601 }
1602
1603 int bpf_object__unload(struct bpf_object *obj)
1604 {
1605         size_t i;
1606
1607         if (!obj)
1608                 return -EINVAL;
1609
1610         for (i = 0; i < obj->nr_maps; i++)
1611                 zclose(obj->maps[i].fd);
1612
1613         for (i = 0; i < obj->nr_programs; i++)
1614                 bpf_program__unload(&obj->programs[i]);
1615
1616         return 0;
1617 }
1618
1619 int bpf_object__load(struct bpf_object *obj)
1620 {
1621         int err;
1622
1623         if (!obj)
1624                 return -EINVAL;
1625
1626         if (obj->loaded) {
1627                 pr_warning("object should not be loaded twice\n");
1628                 return -EINVAL;
1629         }
1630
1631         obj->loaded = true;
1632
1633         CHECK_ERR(bpf_object__create_maps(obj), err, out);
1634         CHECK_ERR(bpf_object__relocate(obj), err, out);
1635         CHECK_ERR(bpf_object__load_progs(obj), err, out);
1636
1637         return 0;
1638 out:
1639         bpf_object__unload(obj);
1640         pr_warning("failed to load object '%s'\n", obj->path);
1641         return err;
1642 }
1643
1644 static int check_path(const char *path)
1645 {
1646         char *cp, errmsg[STRERR_BUFSIZE];
1647         struct statfs st_fs;
1648         char *dname, *dir;
1649         int err = 0;
1650
1651         if (path == NULL)
1652                 return -EINVAL;
1653
1654         dname = strdup(path);
1655         if (dname == NULL)
1656                 return -ENOMEM;
1657
1658         dir = dirname(dname);
1659         if (statfs(dir, &st_fs)) {
1660                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1661                 pr_warning("failed to statfs %s: %s\n", dir, cp);
1662                 err = -errno;
1663         }
1664         free(dname);
1665
1666         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
1667                 pr_warning("specified path %s is not on BPF FS\n", path);
1668                 err = -EINVAL;
1669         }
1670
1671         return err;
1672 }
1673
1674 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
1675                               int instance)
1676 {
1677         char *cp, errmsg[STRERR_BUFSIZE];
1678         int err;
1679
1680         err = check_path(path);
1681         if (err)
1682                 return err;
1683
1684         if (prog == NULL) {
1685                 pr_warning("invalid program pointer\n");
1686                 return -EINVAL;
1687         }
1688
1689         if (instance < 0 || instance >= prog->instances.nr) {
1690                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
1691                            instance, prog->section_name, prog->instances.nr);
1692                 return -EINVAL;
1693         }
1694
1695         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
1696                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1697                 pr_warning("failed to pin program: %s\n", cp);
1698                 return -errno;
1699         }
1700         pr_debug("pinned program '%s'\n", path);
1701
1702         return 0;
1703 }
1704
1705 static int make_dir(const char *path)
1706 {
1707         char *cp, errmsg[STRERR_BUFSIZE];
1708         int err = 0;
1709
1710         if (mkdir(path, 0700) && errno != EEXIST)
1711                 err = -errno;
1712
1713         if (err) {
1714                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
1715                 pr_warning("failed to mkdir %s: %s\n", path, cp);
1716         }
1717         return err;
1718 }
1719
1720 int bpf_program__pin(struct bpf_program *prog, const char *path)
1721 {
1722         int i, err;
1723
1724         err = check_path(path);
1725         if (err)
1726                 return err;
1727
1728         if (prog == NULL) {
1729                 pr_warning("invalid program pointer\n");
1730                 return -EINVAL;
1731         }
1732
1733         if (prog->instances.nr <= 0) {
1734                 pr_warning("no instances of prog %s to pin\n",
1735                            prog->section_name);
1736                 return -EINVAL;
1737         }
1738
1739         err = make_dir(path);
1740         if (err)
1741                 return err;
1742
1743         for (i = 0; i < prog->instances.nr; i++) {
1744                 char buf[PATH_MAX];
1745                 int len;
1746
1747                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
1748                 if (len < 0)
1749                         return -EINVAL;
1750                 else if (len >= PATH_MAX)
1751                         return -ENAMETOOLONG;
1752
1753                 err = bpf_program__pin_instance(prog, buf, i);
1754                 if (err)
1755                         return err;
1756         }
1757
1758         return 0;
1759 }
1760
1761 int bpf_map__pin(struct bpf_map *map, const char *path)
1762 {
1763         char *cp, errmsg[STRERR_BUFSIZE];
1764         int err;
1765
1766         err = check_path(path);
1767         if (err)
1768                 return err;
1769
1770         if (map == NULL) {
1771                 pr_warning("invalid map pointer\n");
1772                 return -EINVAL;
1773         }
1774
1775         if (bpf_obj_pin(map->fd, path)) {
1776                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
1777                 pr_warning("failed to pin map: %s\n", cp);
1778                 return -errno;
1779         }
1780
1781         pr_debug("pinned map '%s'\n", path);
1782         return 0;
1783 }
1784
1785 int bpf_object__pin(struct bpf_object *obj, const char *path)
1786 {
1787         struct bpf_program *prog;
1788         struct bpf_map *map;
1789         int err;
1790
1791         if (!obj)
1792                 return -ENOENT;
1793
1794         if (!obj->loaded) {
1795                 pr_warning("object not yet loaded; load it first\n");
1796                 return -ENOENT;
1797         }
1798
1799         err = make_dir(path);
1800         if (err)
1801                 return err;
1802
1803         bpf_map__for_each(map, obj) {
1804                 char buf[PATH_MAX];
1805                 int len;
1806
1807                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1808                                bpf_map__name(map));
1809                 if (len < 0)
1810                         return -EINVAL;
1811                 else if (len >= PATH_MAX)
1812                         return -ENAMETOOLONG;
1813
1814                 err = bpf_map__pin(map, buf);
1815                 if (err)
1816                         return err;
1817         }
1818
1819         bpf_object__for_each_program(prog, obj) {
1820                 char buf[PATH_MAX];
1821                 int len;
1822
1823                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1824                                prog->section_name);
1825                 if (len < 0)
1826                         return -EINVAL;
1827                 else if (len >= PATH_MAX)
1828                         return -ENAMETOOLONG;
1829
1830                 err = bpf_program__pin(prog, buf);
1831                 if (err)
1832                         return err;
1833         }
1834
1835         return 0;
1836 }
1837
1838 void bpf_object__close(struct bpf_object *obj)
1839 {
1840         size_t i;
1841
1842         if (!obj)
1843                 return;
1844
1845         if (obj->clear_priv)
1846                 obj->clear_priv(obj, obj->priv);
1847
1848         bpf_object__elf_finish(obj);
1849         bpf_object__unload(obj);
1850         btf__free(obj->btf);
1851
1852         for (i = 0; i < obj->nr_maps; i++) {
1853                 zfree(&obj->maps[i].name);
1854                 if (obj->maps[i].clear_priv)
1855                         obj->maps[i].clear_priv(&obj->maps[i],
1856                                                 obj->maps[i].priv);
1857                 obj->maps[i].priv = NULL;
1858                 obj->maps[i].clear_priv = NULL;
1859         }
1860         zfree(&obj->maps);
1861         obj->nr_maps = 0;
1862
1863         if (obj->programs && obj->nr_programs) {
1864                 for (i = 0; i < obj->nr_programs; i++)
1865                         bpf_program__exit(&obj->programs[i]);
1866         }
1867         zfree(&obj->programs);
1868
1869         list_del(&obj->list);
1870         free(obj);
1871 }
1872
1873 struct bpf_object *
1874 bpf_object__next(struct bpf_object *prev)
1875 {
1876         struct bpf_object *next;
1877
1878         if (!prev)
1879                 next = list_first_entry(&bpf_objects_list,
1880                                         struct bpf_object,
1881                                         list);
1882         else
1883                 next = list_next_entry(prev, list);
1884
1885         /* Empty list is noticed here so don't need checking on entry. */
1886         if (&next->list == &bpf_objects_list)
1887                 return NULL;
1888
1889         return next;
1890 }
1891
1892 const char *bpf_object__name(struct bpf_object *obj)
1893 {
1894         return obj ? obj->path : ERR_PTR(-EINVAL);
1895 }
1896
1897 unsigned int bpf_object__kversion(struct bpf_object *obj)
1898 {
1899         return obj ? obj->kern_version : 0;
1900 }
1901
1902 int bpf_object__btf_fd(const struct bpf_object *obj)
1903 {
1904         return obj->btf ? btf__fd(obj->btf) : -1;
1905 }
1906
1907 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
1908                          bpf_object_clear_priv_t clear_priv)
1909 {
1910         if (obj->priv && obj->clear_priv)
1911                 obj->clear_priv(obj, obj->priv);
1912
1913         obj->priv = priv;
1914         obj->clear_priv = clear_priv;
1915         return 0;
1916 }
1917
1918 void *bpf_object__priv(struct bpf_object *obj)
1919 {
1920         return obj ? obj->priv : ERR_PTR(-EINVAL);
1921 }
1922
1923 static struct bpf_program *
1924 __bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1925 {
1926         size_t idx;
1927
1928         if (!obj->programs)
1929                 return NULL;
1930         /* First handler */
1931         if (prev == NULL)
1932                 return &obj->programs[0];
1933
1934         if (prev->obj != obj) {
1935                 pr_warning("error: program handler doesn't match object\n");
1936                 return NULL;
1937         }
1938
1939         idx = (prev - obj->programs) + 1;
1940         if (idx >= obj->nr_programs)
1941                 return NULL;
1942         return &obj->programs[idx];
1943 }
1944
1945 struct bpf_program *
1946 bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1947 {
1948         struct bpf_program *prog = prev;
1949
1950         do {
1951                 prog = __bpf_program__next(prog, obj);
1952         } while (prog && bpf_program__is_function_storage(prog, obj));
1953
1954         return prog;
1955 }
1956
1957 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
1958                           bpf_program_clear_priv_t clear_priv)
1959 {
1960         if (prog->priv && prog->clear_priv)
1961                 prog->clear_priv(prog, prog->priv);
1962
1963         prog->priv = priv;
1964         prog->clear_priv = clear_priv;
1965         return 0;
1966 }
1967
1968 void *bpf_program__priv(struct bpf_program *prog)
1969 {
1970         return prog ? prog->priv : ERR_PTR(-EINVAL);
1971 }
1972
1973 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
1974 {
1975         prog->prog_ifindex = ifindex;
1976 }
1977
1978 const char *bpf_program__title(struct bpf_program *prog, bool needs_copy)
1979 {
1980         const char *title;
1981
1982         title = prog->section_name;
1983         if (needs_copy) {
1984                 title = strdup(title);
1985                 if (!title) {
1986                         pr_warning("failed to strdup program title\n");
1987                         return ERR_PTR(-ENOMEM);
1988                 }
1989         }
1990
1991         return title;
1992 }
1993
1994 int bpf_program__fd(struct bpf_program *prog)
1995 {
1996         return bpf_program__nth_fd(prog, 0);
1997 }
1998
1999 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
2000                           bpf_program_prep_t prep)
2001 {
2002         int *instances_fds;
2003
2004         if (nr_instances <= 0 || !prep)
2005                 return -EINVAL;
2006
2007         if (prog->instances.nr > 0 || prog->instances.fds) {
2008                 pr_warning("Can't set pre-processor after loading\n");
2009                 return -EINVAL;
2010         }
2011
2012         instances_fds = malloc(sizeof(int) * nr_instances);
2013         if (!instances_fds) {
2014                 pr_warning("alloc memory failed for fds\n");
2015                 return -ENOMEM;
2016         }
2017
2018         /* fill all fd with -1 */
2019         memset(instances_fds, -1, sizeof(int) * nr_instances);
2020
2021         prog->instances.nr = nr_instances;
2022         prog->instances.fds = instances_fds;
2023         prog->preprocessor = prep;
2024         return 0;
2025 }
2026
2027 int bpf_program__nth_fd(struct bpf_program *prog, int n)
2028 {
2029         int fd;
2030
2031         if (!prog)
2032                 return -EINVAL;
2033
2034         if (n >= prog->instances.nr || n < 0) {
2035                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
2036                            n, prog->section_name, prog->instances.nr);
2037                 return -EINVAL;
2038         }
2039
2040         fd = prog->instances.fds[n];
2041         if (fd < 0) {
2042                 pr_warning("%dth instance of program '%s' is invalid\n",
2043                            n, prog->section_name);
2044                 return -ENOENT;
2045         }
2046
2047         return fd;
2048 }
2049
2050 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
2051 {
2052         prog->type = type;
2053 }
2054
2055 static bool bpf_program__is_type(struct bpf_program *prog,
2056                                  enum bpf_prog_type type)
2057 {
2058         return prog ? (prog->type == type) : false;
2059 }
2060
2061 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                   \
2062 int bpf_program__set_##NAME(struct bpf_program *prog)   \
2063 {                                                       \
2064         if (!prog)                                      \
2065                 return -EINVAL;                         \
2066         bpf_program__set_type(prog, TYPE);              \
2067         return 0;                                       \
2068 }                                                       \
2069                                                         \
2070 bool bpf_program__is_##NAME(struct bpf_program *prog)   \
2071 {                                                       \
2072         return bpf_program__is_type(prog, TYPE);        \
2073 }                                                       \
2074
2075 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
2076 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
2077 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
2078 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
2079 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
2080 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
2081 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
2082 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
2083
2084 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
2085                                            enum bpf_attach_type type)
2086 {
2087         prog->expected_attach_type = type;
2088 }
2089
2090 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, atype) \
2091         { string, sizeof(string) - 1, ptype, eatype, atype }
2092
2093 /* Programs that can NOT be attached. */
2094 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, -EINVAL)
2095
2096 /* Programs that can be attached. */
2097 #define BPF_APROG_SEC(string, ptype, atype) \
2098         BPF_PROG_SEC_IMPL(string, ptype, 0, atype)
2099
2100 /* Programs that must specify expected attach type at load time. */
2101 #define BPF_EAPROG_SEC(string, ptype, eatype) \
2102         BPF_PROG_SEC_IMPL(string, ptype, eatype, eatype)
2103
2104 /* Programs that can be attached but attach type can't be identified by section
2105  * name. Kept for backward compatibility.
2106  */
2107 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
2108
2109 static const struct {
2110         const char *sec;
2111         size_t len;
2112         enum bpf_prog_type prog_type;
2113         enum bpf_attach_type expected_attach_type;
2114         enum bpf_attach_type attach_type;
2115 } section_names[] = {
2116         BPF_PROG_SEC("socket",                  BPF_PROG_TYPE_SOCKET_FILTER),
2117         BPF_PROG_SEC("kprobe/",                 BPF_PROG_TYPE_KPROBE),
2118         BPF_PROG_SEC("kretprobe/",              BPF_PROG_TYPE_KPROBE),
2119         BPF_PROG_SEC("classifier",              BPF_PROG_TYPE_SCHED_CLS),
2120         BPF_PROG_SEC("action",                  BPF_PROG_TYPE_SCHED_ACT),
2121         BPF_PROG_SEC("tracepoint/",             BPF_PROG_TYPE_TRACEPOINT),
2122         BPF_PROG_SEC("raw_tracepoint/",         BPF_PROG_TYPE_RAW_TRACEPOINT),
2123         BPF_PROG_SEC("xdp",                     BPF_PROG_TYPE_XDP),
2124         BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
2125         BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
2126         BPF_PROG_SEC("lwt_out",                 BPF_PROG_TYPE_LWT_OUT),
2127         BPF_PROG_SEC("lwt_xmit",                BPF_PROG_TYPE_LWT_XMIT),
2128         BPF_PROG_SEC("lwt_seg6local",           BPF_PROG_TYPE_LWT_SEG6LOCAL),
2129         BPF_APROG_SEC("cgroup_skb/ingress",     BPF_PROG_TYPE_CGROUP_SKB,
2130                                                 BPF_CGROUP_INET_INGRESS),
2131         BPF_APROG_SEC("cgroup_skb/egress",      BPF_PROG_TYPE_CGROUP_SKB,
2132                                                 BPF_CGROUP_INET_EGRESS),
2133         BPF_APROG_COMPAT("cgroup/skb",          BPF_PROG_TYPE_CGROUP_SKB),
2134         BPF_APROG_SEC("cgroup/sock",            BPF_PROG_TYPE_CGROUP_SOCK,
2135                                                 BPF_CGROUP_INET_SOCK_CREATE),
2136         BPF_EAPROG_SEC("cgroup/post_bind4",     BPF_PROG_TYPE_CGROUP_SOCK,
2137                                                 BPF_CGROUP_INET4_POST_BIND),
2138         BPF_EAPROG_SEC("cgroup/post_bind6",     BPF_PROG_TYPE_CGROUP_SOCK,
2139                                                 BPF_CGROUP_INET6_POST_BIND),
2140         BPF_APROG_SEC("cgroup/dev",             BPF_PROG_TYPE_CGROUP_DEVICE,
2141                                                 BPF_CGROUP_DEVICE),
2142         BPF_APROG_SEC("sockops",                BPF_PROG_TYPE_SOCK_OPS,
2143                                                 BPF_CGROUP_SOCK_OPS),
2144         BPF_APROG_SEC("sk_skb/stream_parser",   BPF_PROG_TYPE_SK_SKB,
2145                                                 BPF_SK_SKB_STREAM_PARSER),
2146         BPF_APROG_SEC("sk_skb/stream_verdict",  BPF_PROG_TYPE_SK_SKB,
2147                                                 BPF_SK_SKB_STREAM_VERDICT),
2148         BPF_APROG_COMPAT("sk_skb",              BPF_PROG_TYPE_SK_SKB),
2149         BPF_APROG_SEC("sk_msg",                 BPF_PROG_TYPE_SK_MSG,
2150                                                 BPF_SK_MSG_VERDICT),
2151         BPF_APROG_SEC("lirc_mode2",             BPF_PROG_TYPE_LIRC_MODE2,
2152                                                 BPF_LIRC_MODE2),
2153         BPF_APROG_SEC("flow_dissector",         BPF_PROG_TYPE_FLOW_DISSECTOR,
2154                                                 BPF_FLOW_DISSECTOR),
2155         BPF_EAPROG_SEC("cgroup/bind4",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
2156                                                 BPF_CGROUP_INET4_BIND),
2157         BPF_EAPROG_SEC("cgroup/bind6",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
2158                                                 BPF_CGROUP_INET6_BIND),
2159         BPF_EAPROG_SEC("cgroup/connect4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
2160                                                 BPF_CGROUP_INET4_CONNECT),
2161         BPF_EAPROG_SEC("cgroup/connect6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
2162                                                 BPF_CGROUP_INET6_CONNECT),
2163         BPF_EAPROG_SEC("cgroup/sendmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
2164                                                 BPF_CGROUP_UDP4_SENDMSG),
2165         BPF_EAPROG_SEC("cgroup/sendmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
2166                                                 BPF_CGROUP_UDP6_SENDMSG),
2167 };
2168
2169 #undef BPF_PROG_SEC_IMPL
2170 #undef BPF_PROG_SEC
2171 #undef BPF_APROG_SEC
2172 #undef BPF_EAPROG_SEC
2173 #undef BPF_APROG_COMPAT
2174
2175 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
2176                              enum bpf_attach_type *expected_attach_type)
2177 {
2178         int i;
2179
2180         if (!name)
2181                 return -EINVAL;
2182
2183         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
2184                 if (strncmp(name, section_names[i].sec, section_names[i].len))
2185                         continue;
2186                 *prog_type = section_names[i].prog_type;
2187                 *expected_attach_type = section_names[i].expected_attach_type;
2188                 return 0;
2189         }
2190         return -EINVAL;
2191 }
2192
2193 int libbpf_attach_type_by_name(const char *name,
2194                                enum bpf_attach_type *attach_type)
2195 {
2196         int i;
2197
2198         if (!name)
2199                 return -EINVAL;
2200
2201         for (i = 0; i < ARRAY_SIZE(section_names); i++) {
2202                 if (strncmp(name, section_names[i].sec, section_names[i].len))
2203                         continue;
2204                 if (section_names[i].attach_type == -EINVAL)
2205                         return -EINVAL;
2206                 *attach_type = section_names[i].attach_type;
2207                 return 0;
2208         }
2209         return -EINVAL;
2210 }
2211
2212 static int
2213 bpf_program__identify_section(struct bpf_program *prog,
2214                               enum bpf_prog_type *prog_type,
2215                               enum bpf_attach_type *expected_attach_type)
2216 {
2217         return libbpf_prog_type_by_name(prog->section_name, prog_type,
2218                                         expected_attach_type);
2219 }
2220
2221 int bpf_map__fd(struct bpf_map *map)
2222 {
2223         return map ? map->fd : -EINVAL;
2224 }
2225
2226 const struct bpf_map_def *bpf_map__def(struct bpf_map *map)
2227 {
2228         return map ? &map->def : ERR_PTR(-EINVAL);
2229 }
2230
2231 const char *bpf_map__name(struct bpf_map *map)
2232 {
2233         return map ? map->name : NULL;
2234 }
2235
2236 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
2237 {
2238         return map ? map->btf_key_type_id : 0;
2239 }
2240
2241 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
2242 {
2243         return map ? map->btf_value_type_id : 0;
2244 }
2245
2246 int bpf_map__set_priv(struct bpf_map *map, void *priv,
2247                      bpf_map_clear_priv_t clear_priv)
2248 {
2249         if (!map)
2250                 return -EINVAL;
2251
2252         if (map->priv) {
2253                 if (map->clear_priv)
2254                         map->clear_priv(map, map->priv);
2255         }
2256
2257         map->priv = priv;
2258         map->clear_priv = clear_priv;
2259         return 0;
2260 }
2261
2262 void *bpf_map__priv(struct bpf_map *map)
2263 {
2264         return map ? map->priv : ERR_PTR(-EINVAL);
2265 }
2266
2267 bool bpf_map__is_offload_neutral(struct bpf_map *map)
2268 {
2269         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
2270 }
2271
2272 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
2273 {
2274         map->map_ifindex = ifindex;
2275 }
2276
2277 struct bpf_map *
2278 bpf_map__next(struct bpf_map *prev, struct bpf_object *obj)
2279 {
2280         size_t idx;
2281         struct bpf_map *s, *e;
2282
2283         if (!obj || !obj->maps)
2284                 return NULL;
2285
2286         s = obj->maps;
2287         e = obj->maps + obj->nr_maps;
2288
2289         if (prev == NULL)
2290                 return s;
2291
2292         if ((prev < s) || (prev >= e)) {
2293                 pr_warning("error in %s: map handler doesn't belong to object\n",
2294                            __func__);
2295                 return NULL;
2296         }
2297
2298         idx = (prev - obj->maps) + 1;
2299         if (idx >= obj->nr_maps)
2300                 return NULL;
2301         return &obj->maps[idx];
2302 }
2303
2304 struct bpf_map *
2305 bpf_object__find_map_by_name(struct bpf_object *obj, const char *name)
2306 {
2307         struct bpf_map *pos;
2308
2309         bpf_map__for_each(pos, obj) {
2310                 if (pos->name && !strcmp(pos->name, name))
2311                         return pos;
2312         }
2313         return NULL;
2314 }
2315
2316 struct bpf_map *
2317 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
2318 {
2319         int i;
2320
2321         for (i = 0; i < obj->nr_maps; i++) {
2322                 if (obj->maps[i].offset == offset)
2323                         return &obj->maps[i];
2324         }
2325         return ERR_PTR(-ENOENT);
2326 }
2327
2328 long libbpf_get_error(const void *ptr)
2329 {
2330         if (IS_ERR(ptr))
2331                 return PTR_ERR(ptr);
2332         return 0;
2333 }
2334
2335 int bpf_prog_load(const char *file, enum bpf_prog_type type,
2336                   struct bpf_object **pobj, int *prog_fd)
2337 {
2338         struct bpf_prog_load_attr attr;
2339
2340         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
2341         attr.file = file;
2342         attr.prog_type = type;
2343         attr.expected_attach_type = 0;
2344
2345         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
2346 }
2347
2348 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
2349                         struct bpf_object **pobj, int *prog_fd)
2350 {
2351         struct bpf_object_open_attr open_attr = {
2352                 .file           = attr->file,
2353                 .prog_type      = attr->prog_type,
2354         };
2355         struct bpf_program *prog, *first_prog = NULL;
2356         enum bpf_attach_type expected_attach_type;
2357         enum bpf_prog_type prog_type;
2358         struct bpf_object *obj;
2359         struct bpf_map *map;
2360         int err;
2361
2362         if (!attr)
2363                 return -EINVAL;
2364         if (!attr->file)
2365                 return -EINVAL;
2366
2367         obj = bpf_object__open_xattr(&open_attr);
2368         if (IS_ERR_OR_NULL(obj))
2369                 return -ENOENT;
2370
2371         bpf_object__for_each_program(prog, obj) {
2372                 /*
2373                  * If type is not specified, try to guess it based on
2374                  * section name.
2375                  */
2376                 prog_type = attr->prog_type;
2377                 prog->prog_ifindex = attr->ifindex;
2378                 expected_attach_type = attr->expected_attach_type;
2379                 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
2380                         err = bpf_program__identify_section(prog, &prog_type,
2381                                                             &expected_attach_type);
2382                         if (err < 0) {
2383                                 pr_warning("failed to guess program type based on section name %s\n",
2384                                            prog->section_name);
2385                                 bpf_object__close(obj);
2386                                 return -EINVAL;
2387                         }
2388                 }
2389
2390                 bpf_program__set_type(prog, prog_type);
2391                 bpf_program__set_expected_attach_type(prog,
2392                                                       expected_attach_type);
2393
2394                 if (!first_prog)
2395                         first_prog = prog;
2396         }
2397
2398         bpf_map__for_each(map, obj) {
2399                 if (!bpf_map__is_offload_neutral(map))
2400                         map->map_ifindex = attr->ifindex;
2401         }
2402
2403         if (!first_prog) {
2404                 pr_warning("object file doesn't contain bpf program\n");
2405                 bpf_object__close(obj);
2406                 return -ENOENT;
2407         }
2408
2409         err = bpf_object__load(obj);
2410         if (err) {
2411                 bpf_object__close(obj);
2412                 return -EINVAL;
2413         }
2414
2415         *pobj = obj;
2416         *prog_fd = bpf_program__fd(first_prog);
2417         return 0;
2418 }
2419
2420 enum bpf_perf_event_ret
2421 bpf_perf_event_read_simple(void *mem, unsigned long size,
2422                            unsigned long page_size, void **buf, size_t *buf_len,
2423                            bpf_perf_event_print_t fn, void *priv)
2424 {
2425         volatile struct perf_event_mmap_page *header = mem;
2426         __u64 data_tail = header->data_tail;
2427         __u64 data_head = header->data_head;
2428         int ret = LIBBPF_PERF_EVENT_ERROR;
2429         void *base, *begin, *end;
2430
2431         asm volatile("" ::: "memory"); /* in real code it should be smp_rmb() */
2432         if (data_head == data_tail)
2433                 return LIBBPF_PERF_EVENT_CONT;
2434
2435         base = ((char *)header) + page_size;
2436
2437         begin = base + data_tail % size;
2438         end = base + data_head % size;
2439
2440         while (begin != end) {
2441                 struct perf_event_header *ehdr;
2442
2443                 ehdr = begin;
2444                 if (begin + ehdr->size > base + size) {
2445                         long len = base + size - begin;
2446
2447                         if (*buf_len < ehdr->size) {
2448                                 free(*buf);
2449                                 *buf = malloc(ehdr->size);
2450                                 if (!*buf) {
2451                                         ret = LIBBPF_PERF_EVENT_ERROR;
2452                                         break;
2453                                 }
2454                                 *buf_len = ehdr->size;
2455                         }
2456
2457                         memcpy(*buf, begin, len);
2458                         memcpy(*buf + len, base, ehdr->size - len);
2459                         ehdr = (void *)*buf;
2460                         begin = base + ehdr->size - len;
2461                 } else if (begin + ehdr->size == base + size) {
2462                         begin = base;
2463                 } else {
2464                         begin += ehdr->size;
2465                 }
2466
2467                 ret = fn(ehdr, priv);
2468                 if (ret != LIBBPF_PERF_EVENT_CONT)
2469                         break;
2470
2471                 data_tail += ehdr->size;
2472         }
2473
2474         __sync_synchronize(); /* smp_mb() */
2475         header->data_tail = data_tail;
2476
2477         return ret;
2478 }