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 #include <stdlib.h>
  26 #include <stdio.h>
  27 #include <stdarg.h>
  28 #include <libgen.h>
  29 #include <inttypes.h>
  30 #include <string.h>
  31 #include <unistd.h>
  32 #include <fcntl.h>
  33 #include <errno.h>
  34 #include <perf-sys.h>
  35 #include <asm/unistd.h>
  36 #include <linux/err.h>
  37 #include <linux/kernel.h>
  38 #include <linux/bpf.h>
  39 #include <linux/btf.h>
  40 #include <linux/list.h>
  41 #include <linux/limits.h>
  42 #include <sys/stat.h>
  43 #include <sys/types.h>
  44 #include <sys/vfs.h>
  45 #include <libelf.h>
  46 #include <gelf.h>
  47
  48 #include "libbpf.h"
  49 #include "bpf.h"
  50 #include "btf.h"
  51
  52 #ifndef EM_BPF
  53 #define EM_BPF 247
  54 #endif
  55
  56 #ifndef BPF_FS_MAGIC
  57 #define BPF_FS_MAGIC            0xcafe4a11
  58 #endif
  59
  60 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  61
  62 __printf(1, 2)
  63 static int __base_pr(const char *format, ...)
  64 {
  65         va_list args;
  66         int err;
  67
  68         va_start(args, format);
  69         err = vfprintf(stderr, format, args);
  70         va_end(args);
  71         return err;
  72 }
  73
  74 static __printf(1, 2) libbpf_print_fn_t __pr_warning = __base_pr;
  75 static __printf(1, 2) libbpf_print_fn_t __pr_info = __base_pr;
  76 static __printf(1, 2) libbpf_print_fn_t __pr_debug;
  77
  78 #define __pr(func, fmt, ...)    \
  79 do {                            \
  80         if ((func))             \
  81                 (func)("libbpf: " fmt, ##__VA_ARGS__); \
  82 } while (0)
  83
  84 #define pr_warning(fmt, ...)    __pr(__pr_warning, fmt, ##__VA_ARGS__)
  85 #define pr_info(fmt, ...)       __pr(__pr_info, fmt, ##__VA_ARGS__)
  86 #define pr_debug(fmt, ...)      __pr(__pr_debug, fmt, ##__VA_ARGS__)
  87
  88 void libbpf_set_print(libbpf_print_fn_t warn,
  89                       libbpf_print_fn_t info,
  90                       libbpf_print_fn_t debug)
  91 {
  92         __pr_warning = warn;
  93         __pr_info = info;
  94         __pr_debug = debug;
  95 }
  96
  97 #define STRERR_BUFSIZE  128
  98
  99 #define ERRNO_OFFSET(e)         ((e) - __LIBBPF_ERRNO__START)
 100 #define ERRCODE_OFFSET(c)       ERRNO_OFFSET(LIBBPF_ERRNO__##c)
 101 #define NR_ERRNO        (__LIBBPF_ERRNO__END - __LIBBPF_ERRNO__START)
 102
 103 static const char *libbpf_strerror_table[NR_ERRNO] = {
 104         [ERRCODE_OFFSET(LIBELF)]        = "Something wrong in libelf",
 105         [ERRCODE_OFFSET(FORMAT)]        = "BPF object format invalid",
 106         [ERRCODE_OFFSET(KVERSION)]      = "'version' section incorrect or lost",
 107         [ERRCODE_OFFSET(ENDIAN)]        = "Endian mismatch",
 108         [ERRCODE_OFFSET(INTERNAL)]      = "Internal error in libbpf",
 109         [ERRCODE_OFFSET(RELOC)]         = "Relocation failed",
 110         [ERRCODE_OFFSET(VERIFY)]        = "Kernel verifier blocks program loading",
 111         [ERRCODE_OFFSET(PROG2BIG)]      = "Program too big",
 112         [ERRCODE_OFFSET(KVER)]          = "Incorrect kernel version",
 113         [ERRCODE_OFFSET(PROGTYPE)]      = "Kernel doesn't support this program type",
 114         [ERRCODE_OFFSET(WRNGPID)]       = "Wrong pid in netlink message",
 115         [ERRCODE_OFFSET(INVSEQ)]        = "Invalid netlink sequence",
 116 };
 117
 118 int libbpf_strerror(int err, char *buf, size_t size)
 119 {
 120         if (!buf || !size)
 121                 return -1;
 122
 123         err = err > 0 ? err : -err;
 124
 125         if (err < __LIBBPF_ERRNO__START) {
 126                 int ret;
 127
 128                 ret = strerror_r(err, buf, size);
 129                 buf[size - 1] = '\0';
 130                 return ret;
 131         }
 132
 133         if (err < __LIBBPF_ERRNO__END) {
 134                 const char *msg;
 135
 136                 msg = libbpf_strerror_table[ERRNO_OFFSET(err)];
 137                 snprintf(buf, size, "%s", msg);
 138                 buf[size - 1] = '\0';
 139                 return 0;
 140         }
 141
 142         snprintf(buf, size, "Unknown libbpf error %d", err);
 143         buf[size - 1] = '\0';
 144         return -1;
 145 }
 146
 147 #define CHECK_ERR(action, err, out) do {        \
 148         err = action;                   \
 149         if (err)                        \
 150                 goto out;               \
 151 } while(0)
 152
 153
 154 /* Copied from tools/perf/util/util.h */
 155 #ifndef zfree
 156 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 157 #endif
 158
 159 #ifndef zclose
 160 # define zclose(fd) ({                  \
 161         int ___err = 0;                 \
 162         if ((fd) >= 0)                  \
 163                 ___err = close((fd));   \
 164         fd = -1;                        \
 165         ___err; })
 166 #endif
 167
 168 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 169 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 170 #else
 171 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 172 #endif
 173
 174 /*
 175  * bpf_prog should be a better name but it has been used in
 176  * linux/filter.h.
 177  */
 178 struct bpf_program {
 179         /* Index in elf obj file, for relocation use. */
 180         int idx;
 181         char *name;
 182         int prog_ifindex;
 183         char *section_name;
 184         struct bpf_insn *insns;
 185         size_t insns_cnt, main_prog_cnt;
 186         enum bpf_prog_type type;
 187
 188         struct reloc_desc {
 189                 enum {
 190                         RELO_LD64,
 191                         RELO_CALL,
 192                 } type;
 193                 int insn_idx;
 194                 union {
 195                         int map_idx;
 196                         int text_off;
 197                 };
 198         } *reloc_desc;
 199         int nr_reloc;
 200
 201         struct {
 202                 int nr;
 203                 int *fds;
 204         } instances;
 205         bpf_program_prep_t preprocessor;
 206
 207         struct bpf_object *obj;
 208         void *priv;
 209         bpf_program_clear_priv_t clear_priv;
 210
 211         enum bpf_attach_type expected_attach_type;
 212 };
 213
 214 struct bpf_map {
 215         int fd;
 216         char *name;
 217         size_t offset;
 218         int map_ifindex;
 219         struct bpf_map_def def;
 220         __u32 btf_key_type_id;
 221         __u32 btf_value_type_id;
 222         void *priv;
 223         bpf_map_clear_priv_t clear_priv;
 224 };
 225
 226 static LIST_HEAD(bpf_objects_list);
 227
 228 struct bpf_object {
 229         char license[64];
 230         u32 kern_version;
 231
 232         struct bpf_program *programs;
 233         size_t nr_programs;
 234         struct bpf_map *maps;
 235         size_t nr_maps;
 236
 237         bool loaded;
 238
 239         /*
 240          * Information when doing elf related work. Only valid if fd
 241          * is valid.
 242          */
 243         struct {
 244                 int fd;
 245                 void *obj_buf;
 246                 size_t obj_buf_sz;
 247                 Elf *elf;
 248                 GElf_Ehdr ehdr;
 249                 Elf_Data *symbols;
 250                 size_t strtabidx;
 251                 struct {
 252                         GElf_Shdr shdr;
 253                         Elf_Data *data;
 254                 } *reloc;
 255                 int nr_reloc;
 256                 int maps_shndx;
 257                 int text_shndx;
 258         } efile;
 259         /*
 260          * All loaded bpf_object is linked in a list, which is
 261          * hidden to caller. bpf_objects__<func> handlers deal with
 262          * all objects.
 263          */
 264         struct list_head list;
 265
 266         struct btf *btf;
 267
 268         void *priv;
 269         bpf_object_clear_priv_t clear_priv;
 270
 271         char path[];
 272 };
 273 #define obj_elf_valid(o)        ((o)->efile.elf)
 274
 275 static void bpf_program__unload(struct bpf_program *prog)
 276 {
 277         int i;
 278
 279         if (!prog)
 280                 return;
 281
 282         /*
 283          * If the object is opened but the program was never loaded,
 284          * it is possible that prog->instances.nr == -1.
 285          */
 286         if (prog->instances.nr > 0) {
 287                 for (i = 0; i < prog->instances.nr; i++)
 288                         zclose(prog->instances.fds[i]);
 289         } else if (prog->instances.nr != -1) {
 290                 pr_warning("Internal error: instances.nr is %d\n",
 291                            prog->instances.nr);
 292         }
 293
 294         prog->instances.nr = -1;
 295         zfree(&prog->instances.fds);
 296 }
 297
 298 static void bpf_program__exit(struct bpf_program *prog)
 299 {
 300         if (!prog)
 301                 return;
 302
 303         if (prog->clear_priv)
 304                 prog->clear_priv(prog, prog->priv);
 305
 306         prog->priv = NULL;
 307         prog->clear_priv = NULL;
 308
 309         bpf_program__unload(prog);
 310         zfree(&prog->name);
 311         zfree(&prog->section_name);
 312         zfree(&prog->insns);
 313         zfree(&prog->reloc_desc);
 314
 315         prog->nr_reloc = 0;
 316         prog->insns_cnt = 0;
 317         prog->idx = -1;
 318 }
 319
 320 static int
 321 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 322                   struct bpf_program *prog)
 323 {
 324         if (size < sizeof(struct bpf_insn)) {
 325                 pr_warning("corrupted section '%s'\n", section_name);
 326                 return -EINVAL;
 327         }
 328
 329         bzero(prog, sizeof(*prog));
 330
 331         prog->section_name = strdup(section_name);
 332         if (!prog->section_name) {
 333                 pr_warning("failed to alloc name for prog under section(%d) %s\n",
 334                            idx, section_name);
 335                 goto errout;
 336         }
 337
 338         prog->insns = malloc(size);
 339         if (!prog->insns) {
 340                 pr_warning("failed to alloc insns for prog under section %s\n",
 341                            section_name);
 342                 goto errout;
 343         }
 344         prog->insns_cnt = size / sizeof(struct bpf_insn);
 345         memcpy(prog->insns, data,
 346                prog->insns_cnt * sizeof(struct bpf_insn));
 347         prog->idx = idx;
 348         prog->instances.fds = NULL;
 349         prog->instances.nr = -1;
 350         prog->type = BPF_PROG_TYPE_KPROBE;
 351
 352         return 0;
 353 errout:
 354         bpf_program__exit(prog);
 355         return -ENOMEM;
 356 }
 357
 358 static int
 359 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 360                         char *section_name, int idx)
 361 {
 362         struct bpf_program prog, *progs;
 363         int nr_progs, err;
 364
 365         err = bpf_program__init(data, size, section_name, idx, &prog);
 366         if (err)
 367                 return err;
 368
 369         progs = obj->programs;
 370         nr_progs = obj->nr_programs;
 371
 372         progs = realloc(progs, sizeof(progs[0]) * (nr_progs + 1));
 373         if (!progs) {
 374                 /*
 375                  * In this case the original obj->programs
 376                  * is still valid, so don't need special treat for
 377                  * bpf_close_object().
 378                  */
 379                 pr_warning("failed to alloc a new program under section '%s'\n",
 380                            section_name);
 381                 bpf_program__exit(&prog);
 382                 return -ENOMEM;
 383         }
 384
 385         pr_debug("found program %s\n", prog.section_name);
 386         obj->programs = progs;
 387         obj->nr_programs = nr_progs + 1;
 388         prog.obj = obj;
 389         progs[nr_progs] = prog;
 390         return 0;
 391 }
 392
 393 static int
 394 bpf_object__init_prog_names(struct bpf_object *obj)
 395 {
 396         Elf_Data *symbols = obj->efile.symbols;
 397         struct bpf_program *prog;
 398         size_t pi, si;
 399
 400         for (pi = 0; pi < obj->nr_programs; pi++) {
 401                 const char *name = NULL;
 402
 403                 prog = &obj->programs[pi];
 404                 if (prog->idx == obj->efile.text_shndx) {
 405                         name = ".text";
 406                         goto skip_search;
 407                 }
 408
 409                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 410                      si++) {
 411                         GElf_Sym sym;
 412
 413                         if (!gelf_getsym(symbols, si, &sym))
 414                                 continue;
 415                         if (sym.st_shndx != prog->idx)
 416                                 continue;
 417                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 418                                 continue;
 419
 420                         name = elf_strptr(obj->efile.elf,
 421                                           obj->efile.strtabidx,
 422                                           sym.st_name);
 423                         if (!name) {
 424                                 pr_warning("failed to get sym name string for prog %s\n",
 425                                            prog->section_name);
 426                                 return -LIBBPF_ERRNO__LIBELF;
 427                         }
 428                 }
 429
 430                 if (!name) {
 431                         pr_warning("failed to find sym for prog %s\n",
 432                                    prog->section_name);
 433                         return -EINVAL;
 434                 }
 435 skip_search:
 436                 prog->name = strdup(name);
 437                 if (!prog->name) {
 438                         pr_warning("failed to allocate memory for prog sym %s\n",
 439                                    name);
 440                         return -ENOMEM;
 441                 }
 442         }
 443
 444         return 0;
 445 }
 446
 447 static struct bpf_object *bpf_object__new(const char *path,
 448                                           void *obj_buf,
 449                                           size_t obj_buf_sz)
 450 {
 451         struct bpf_object *obj;
 452
 453         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 454         if (!obj) {
 455                 pr_warning("alloc memory failed for %s\n", path);
 456                 return ERR_PTR(-ENOMEM);
 457         }
 458
 459         strcpy(obj->path, path);
 460         obj->efile.fd = -1;
 461
 462         /*
 463          * Caller of this function should also calls
 464          * bpf_object__elf_finish() after data collection to return
 465          * obj_buf to user. If not, we should duplicate the buffer to
 466          * avoid user freeing them before elf finish.
 467          */
 468         obj->efile.obj_buf = obj_buf;
 469         obj->efile.obj_buf_sz = obj_buf_sz;
 470         obj->efile.maps_shndx = -1;
 471
 472         obj->loaded = false;
 473
 474         INIT_LIST_HEAD(&obj->list);
 475         list_add(&obj->list, &bpf_objects_list);
 476         return obj;
 477 }
 478
 479 static void bpf_object__elf_finish(struct bpf_object *obj)
 480 {
 481         if (!obj_elf_valid(obj))
 482                 return;
 483
 484         if (obj->efile.elf) {
 485                 elf_end(obj->efile.elf);
 486                 obj->efile.elf = NULL;
 487         }
 488         obj->efile.symbols = NULL;
 489
 490         zfree(&obj->efile.reloc);
 491         obj->efile.nr_reloc = 0;
 492         zclose(obj->efile.fd);
 493         obj->efile.obj_buf = NULL;
 494         obj->efile.obj_buf_sz = 0;
 495 }
 496
 497 static int bpf_object__elf_init(struct bpf_object *obj)
 498 {
 499         int err = 0;
 500         GElf_Ehdr *ep;
 501
 502         if (obj_elf_valid(obj)) {
 503                 pr_warning("elf init: internal error\n");
 504                 return -LIBBPF_ERRNO__LIBELF;
 505         }
 506
 507         if (obj->efile.obj_buf_sz > 0) {
 508                 /*
 509                  * obj_buf should have been validated by
 510                  * bpf_object__open_buffer().
 511                  */
 512                 obj->efile.elf = elf_memory(obj->efile.obj_buf,
 513                                             obj->efile.obj_buf_sz);
 514         } else {
 515                 obj->efile.fd = open(obj->path, O_RDONLY);
 516                 if (obj->efile.fd < 0) {
 517                         pr_warning("failed to open %s: %s\n", obj->path,
 518                                         strerror(errno));
 519                         return -errno;
 520                 }
 521
 522                 obj->efile.elf = elf_begin(obj->efile.fd,
 523                                 LIBBPF_ELF_C_READ_MMAP,
 524                                 NULL);
 525         }
 526
 527         if (!obj->efile.elf) {
 528                 pr_warning("failed to open %s as ELF file\n",
 529                                 obj->path);
 530                 err = -LIBBPF_ERRNO__LIBELF;
 531                 goto errout;
 532         }
 533
 534         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 535                 pr_warning("failed to get EHDR from %s\n",
 536                                 obj->path);
 537                 err = -LIBBPF_ERRNO__FORMAT;
 538                 goto errout;
 539         }
 540         ep = &obj->efile.ehdr;
 541
 542         /* Old LLVM set e_machine to EM_NONE */
 543         if ((ep->e_type != ET_REL) || (ep->e_machine && (ep->e_machine != EM_BPF))) {
 544                 pr_warning("%s is not an eBPF object file\n",
 545                         obj->path);
 546                 err = -LIBBPF_ERRNO__FORMAT;
 547                 goto errout;
 548         }
 549
 550         return 0;
 551 errout:
 552         bpf_object__elf_finish(obj);
 553         return err;
 554 }
 555
 556 static int
 557 bpf_object__check_endianness(struct bpf_object *obj)
 558 {
 559         static unsigned int const endian = 1;
 560
 561         switch (obj->efile.ehdr.e_ident[EI_DATA]) {
 562         case ELFDATA2LSB:
 563                 /* We are big endian, BPF obj is little endian. */
 564                 if (*(unsigned char const *)&endian != 1)
 565                         goto mismatch;
 566                 break;
 567
 568         case ELFDATA2MSB:
 569                 /* We are little endian, BPF obj is big endian. */
 570                 if (*(unsigned char const *)&endian != 0)
 571                         goto mismatch;
 572                 break;
 573         default:
 574                 return -LIBBPF_ERRNO__ENDIAN;
 575         }
 576
 577         return 0;
 578
 579 mismatch:
 580         pr_warning("Error: endianness mismatch.\n");
 581         return -LIBBPF_ERRNO__ENDIAN;
 582 }
 583
 584 static int
 585 bpf_object__init_license(struct bpf_object *obj,
 586                          void *data, size_t size)
 587 {
 588         memcpy(obj->license, data,
 589                min(size, sizeof(obj->license) - 1));
 590         pr_debug("license of %s is %s\n", obj->path, obj->license);
 591         return 0;
 592 }
 593
 594 static int
 595 bpf_object__init_kversion(struct bpf_object *obj,
 596                           void *data, size_t size)
 597 {
 598         u32 kver;
 599
 600         if (size != sizeof(kver)) {
 601                 pr_warning("invalid kver section in %s\n", obj->path);
 602                 return -LIBBPF_ERRNO__FORMAT;
 603         }
 604         memcpy(&kver, data, sizeof(kver));
 605         obj->kern_version = kver;
 606         pr_debug("kernel version of %s is %x\n", obj->path,
 607                  obj->kern_version);
 608         return 0;
 609 }
 610
 611 static int compare_bpf_map(const void *_a, const void *_b)
 612 {
 613         const struct bpf_map *a = _a;
 614         const struct bpf_map *b = _b;
 615
 616         return a->offset - b->offset;
 617 }
 618
 619 static int
 620 bpf_object__init_maps(struct bpf_object *obj)
 621 {
 622         int i, map_idx, map_def_sz, nr_maps = 0;
 623         Elf_Scn *scn;
 624         Elf_Data *data;
 625         Elf_Data *symbols = obj->efile.symbols;
 626
 627         if (obj->efile.maps_shndx < 0)
 628                 return -EINVAL;
 629         if (!symbols)
 630                 return -EINVAL;
 631
 632         scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
 633         if (scn)
 634                 data = elf_getdata(scn, NULL);
 635         if (!scn || !data) {
 636                 pr_warning("failed to get Elf_Data from map section %d\n",
 637                            obj->efile.maps_shndx);
 638                 return -EINVAL;
 639         }
 640
 641         /*
 642          * Count number of maps. Each map has a name.
 643          * Array of maps is not supported: only the first element is
 644          * considered.
 645          *
 646          * TODO: Detect array of map and report error.
 647          */
 648         for (i = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 649                 GElf_Sym sym;
 650
 651                 if (!gelf_getsym(symbols, i, &sym))
 652                         continue;
 653                 if (sym.st_shndx != obj->efile.maps_shndx)
 654                         continue;
 655                 nr_maps++;
 656         }
 657
 658         /* Alloc obj->maps and fill nr_maps. */
 659         pr_debug("maps in %s: %d maps in %zd bytes\n", obj->path,
 660                  nr_maps, data->d_size);
 661
 662         if (!nr_maps)
 663                 return 0;
 664
 665         /* Assume equally sized map definitions */
 666         map_def_sz = data->d_size / nr_maps;
 667         if (!data->d_size || (data->d_size % nr_maps) != 0) {
 668                 pr_warning("unable to determine map definition size "
 669                            "section %s, %d maps in %zd bytes\n",
 670                            obj->path, nr_maps, data->d_size);
 671                 return -EINVAL;
 672         }
 673
 674         obj->maps = calloc(nr_maps, sizeof(obj->maps[0]));
 675         if (!obj->maps) {
 676                 pr_warning("alloc maps for object failed\n");
 677                 return -ENOMEM;
 678         }
 679         obj->nr_maps = nr_maps;
 680
 681         /*
 682          * fill all fd with -1 so won't close incorrect
 683          * fd (fd=0 is stdin) when failure (zclose won't close
 684          * negative fd)).
 685          */
 686         for (i = 0; i < nr_maps; i++)
 687                 obj->maps[i].fd = -1;
 688
 689         /*
 690          * Fill obj->maps using data in "maps" section.
 691          */
 692         for (i = 0, map_idx = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 693                 GElf_Sym sym;
 694                 const char *map_name;
 695                 struct bpf_map_def *def;
 696
 697                 if (!gelf_getsym(symbols, i, &sym))
 698                         continue;
 699                 if (sym.st_shndx != obj->efile.maps_shndx)
 700                         continue;
 701
 702                 map_name = elf_strptr(obj->efile.elf,
 703                                       obj->efile.strtabidx,
 704                                       sym.st_name);
 705                 obj->maps[map_idx].offset = sym.st_value;
 706                 if (sym.st_value + map_def_sz > data->d_size) {
 707                         pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
 708                                    obj->path, map_name);
 709                         return -EINVAL;
 710                 }
 711
 712                 obj->maps[map_idx].name = strdup(map_name);
 713                 if (!obj->maps[map_idx].name) {
 714                         pr_warning("failed to alloc map name\n");
 715                         return -ENOMEM;
 716                 }
 717                 pr_debug("map %d is \"%s\"\n", map_idx,
 718                          obj->maps[map_idx].name);
 719                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
 720                 /*
 721                  * If the definition of the map in the object file fits in
 722                  * bpf_map_def, copy it.  Any extra fields in our version
 723                  * of bpf_map_def will default to zero as a result of the
 724                  * calloc above.
 725                  */
 726                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
 727                         memcpy(&obj->maps[map_idx].def, def, map_def_sz);
 728                 } else {
 729                         /*
 730                          * Here the map structure being read is bigger than what
 731                          * we expect, truncate if the excess bits are all zero.
 732                          * If they are not zero, reject this map as
 733                          * incompatible.
 734                          */
 735                         char *b;
 736                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
 737                              b < ((char *)def) + map_def_sz; b++) {
 738                                 if (*b != 0) {
 739                                         pr_warning("maps section in %s: \"%s\" "
 740                                                    "has unrecognized, non-zero "
 741                                                    "options\n",
 742                                                    obj->path, map_name);
 743                                         return -EINVAL;
 744                                 }
 745                         }
 746                         memcpy(&obj->maps[map_idx].def, def,
 747                                sizeof(struct bpf_map_def));
 748                 }
 749                 map_idx++;
 750         }
 751
 752         qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]), compare_bpf_map);
 753         return 0;
 754 }
 755
 756 static bool section_have_execinstr(struct bpf_object *obj, int idx)
 757 {
 758         Elf_Scn *scn;
 759         GElf_Shdr sh;
 760
 761         scn = elf_getscn(obj->efile.elf, idx);
 762         if (!scn)
 763                 return false;
 764
 765         if (gelf_getshdr(scn, &sh) != &sh)
 766                 return false;
 767
 768         if (sh.sh_flags & SHF_EXECINSTR)
 769                 return true;
 770
 771         return false;
 772 }
 773
 774 static int bpf_object__elf_collect(struct bpf_object *obj)
 775 {
 776         Elf *elf = obj->efile.elf;
 777         GElf_Ehdr *ep = &obj->efile.ehdr;
 778         Elf_Scn *scn = NULL;
 779         int idx = 0, err = 0;
 780
 781         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
 782         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
 783                 pr_warning("failed to get e_shstrndx from %s\n",
 784                            obj->path);
 785                 return -LIBBPF_ERRNO__FORMAT;
 786         }
 787
 788         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 789                 char *name;
 790                 GElf_Shdr sh;
 791                 Elf_Data *data;
 792
 793                 idx++;
 794                 if (gelf_getshdr(scn, &sh) != &sh) {
 795                         pr_warning("failed to get section(%d) header from %s\n",
 796                                    idx, obj->path);
 797                         err = -LIBBPF_ERRNO__FORMAT;
 798                         goto out;
 799                 }
 800
 801                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 802                 if (!name) {
 803                         pr_warning("failed to get section(%d) name from %s\n",
 804                                    idx, obj->path);
 805                         err = -LIBBPF_ERRNO__FORMAT;
 806                         goto out;
 807                 }
 808
 809                 data = elf_getdata(scn, 0);
 810                 if (!data) {
 811                         pr_warning("failed to get section(%d) data from %s(%s)\n",
 812                                    idx, name, obj->path);
 813                         err = -LIBBPF_ERRNO__FORMAT;
 814                         goto out;
 815                 }
 816                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
 817                          idx, name, (unsigned long)data->d_size,
 818                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
 819                          (int)sh.sh_type);
 820
 821                 if (strcmp(name, "license") == 0)
 822                         err = bpf_object__init_license(obj,
 823                                                        data->d_buf,
 824                                                        data->d_size);
 825                 else if (strcmp(name, "version") == 0)
 826                         err = bpf_object__init_kversion(obj,
 827                                                         data->d_buf,
 828                                                         data->d_size);
 829                 else if (strcmp(name, "maps") == 0)
 830                         obj->efile.maps_shndx = idx;
 831                 else if (strcmp(name, BTF_ELF_SEC) == 0) {
 832                         obj->btf = btf__new(data->d_buf, data->d_size,
 833                                             __pr_debug);
 834                         if (IS_ERR(obj->btf)) {
 835                                 pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
 836                                            BTF_ELF_SEC, PTR_ERR(obj->btf));
 837                                 obj->btf = NULL;
 838                         }
 839                 } else if (sh.sh_type == SHT_SYMTAB) {
 840                         if (obj->efile.symbols) {
 841                                 pr_warning("bpf: multiple SYMTAB in %s\n",
 842                                            obj->path);
 843                                 err = -LIBBPF_ERRNO__FORMAT;
 844                         } else {
 845                                 obj->efile.symbols = data;
 846                                 obj->efile.strtabidx = sh.sh_link;
 847                         }
 848                 } else if ((sh.sh_type == SHT_PROGBITS) &&
 849                            (sh.sh_flags & SHF_EXECINSTR) &&
 850                            (data->d_size > 0)) {
 851                         if (strcmp(name, ".text") == 0)
 852                                 obj->efile.text_shndx = idx;
 853                         err = bpf_object__add_program(obj, data->d_buf,
 854                                                       data->d_size, name, idx);
 855                         if (err) {
 856                                 char errmsg[STRERR_BUFSIZE];
 857
 858                                 strerror_r(-err, errmsg, sizeof(errmsg));
 859                                 pr_warning("failed to alloc program %s (%s): %s",
 860                                            name, obj->path, errmsg);
 861                         }
 862                 } else if (sh.sh_type == SHT_REL) {
 863                         void *reloc = obj->efile.reloc;
 864                         int nr_reloc = obj->efile.nr_reloc + 1;
 865                         int sec = sh.sh_info; /* points to other section */
 866
 867                         /* Only do relo for section with exec instructions */
 868                         if (!section_have_execinstr(obj, sec)) {
 869                                 pr_debug("skip relo %s(%d) for section(%d)\n",
 870                                          name, idx, sec);
 871                                 continue;
 872                         }
 873
 874                         reloc = realloc(reloc,
 875                                         sizeof(*obj->efile.reloc) * nr_reloc);
 876                         if (!reloc) {
 877                                 pr_warning("realloc failed\n");
 878                                 err = -ENOMEM;
 879                         } else {
 880                                 int n = nr_reloc - 1;
 881
 882                                 obj->efile.reloc = reloc;
 883                                 obj->efile.nr_reloc = nr_reloc;
 884
 885                                 obj->efile.reloc[n].shdr = sh;
 886                                 obj->efile.reloc[n].data = data;
 887                         }
 888                 } else {
 889                         pr_debug("skip section(%d) %s\n", idx, name);
 890                 }
 891                 if (err)
 892                         goto out;
 893         }
 894
 895         if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
 896                 pr_warning("Corrupted ELF file: index of strtab invalid\n");
 897                 return LIBBPF_ERRNO__FORMAT;
 898         }
 899         if (obj->efile.maps_shndx >= 0) {
 900                 err = bpf_object__init_maps(obj);
 901                 if (err)
 902                         goto out;
 903         }
 904         err = bpf_object__init_prog_names(obj);
 905 out:
 906         return err;
 907 }
 908
 909 static struct bpf_program *
 910 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
 911 {
 912         struct bpf_program *prog;
 913         size_t i;
 914
 915         for (i = 0; i < obj->nr_programs; i++) {
 916                 prog = &obj->programs[i];
 917                 if (prog->idx == idx)
 918                         return prog;
 919         }
 920         return NULL;
 921 }
 922
 923 static int
 924 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
 925                            Elf_Data *data, struct bpf_object *obj)
 926 {
 927         Elf_Data *symbols = obj->efile.symbols;
 928         int text_shndx = obj->efile.text_shndx;
 929         int maps_shndx = obj->efile.maps_shndx;
 930         struct bpf_map *maps = obj->maps;
 931         size_t nr_maps = obj->nr_maps;
 932         int i, nrels;
 933
 934         pr_debug("collecting relocating info for: '%s'\n",
 935                  prog->section_name);
 936         nrels = shdr->sh_size / shdr->sh_entsize;
 937
 938         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
 939         if (!prog->reloc_desc) {
 940                 pr_warning("failed to alloc memory in relocation\n");
 941                 return -ENOMEM;
 942         }
 943         prog->nr_reloc = nrels;
 944
 945         for (i = 0; i < nrels; i++) {
 946                 GElf_Sym sym;
 947                 GElf_Rel rel;
 948                 unsigned int insn_idx;
 949                 struct bpf_insn *insns = prog->insns;
 950                 size_t map_idx;
 951
 952                 if (!gelf_getrel(data, i, &rel)) {
 953                         pr_warning("relocation: failed to get %d reloc\n", i);
 954                         return -LIBBPF_ERRNO__FORMAT;
 955                 }
 956
 957                 if (!gelf_getsym(symbols,
 958                                  GELF_R_SYM(rel.r_info),
 959                                  &sym)) {
 960                         pr_warning("relocation: symbol %"PRIx64" not found\n",
 961                                    GELF_R_SYM(rel.r_info));
 962                         return -LIBBPF_ERRNO__FORMAT;
 963                 }
 964                 pr_debug("relo for %lld value %lld name %d\n",
 965                          (long long) (rel.r_info >> 32),
 966                          (long long) sym.st_value, sym.st_name);
 967
 968                 if (sym.st_shndx != maps_shndx && sym.st_shndx != text_shndx) {
 969                         pr_warning("Program '%s' contains non-map related relo data pointing to section %u\n",
 970                                    prog->section_name, sym.st_shndx);
 971                         return -LIBBPF_ERRNO__RELOC;
 972                 }
 973
 974                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
 975                 pr_debug("relocation: insn_idx=%u\n", insn_idx);
 976
 977                 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
 978                         if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
 979                                 pr_warning("incorrect bpf_call opcode\n");
 980                                 return -LIBBPF_ERRNO__RELOC;
 981                         }
 982                         prog->reloc_desc[i].type = RELO_CALL;
 983                         prog->reloc_desc[i].insn_idx = insn_idx;
 984                         prog->reloc_desc[i].text_off = sym.st_value;
 985                         continue;
 986                 }
 987
 988                 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
 989                         pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
 990                                    insn_idx, insns[insn_idx].code);
 991                         return -LIBBPF_ERRNO__RELOC;
 992                 }
 993
 994                 /* TODO: 'maps' is sorted. We can use bsearch to make it faster. */
 995                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
 996                         if (maps[map_idx].offset == sym.st_value) {
 997                                 pr_debug("relocation: find map %zd (%s) for insn %u\n",
 998                                          map_idx, maps[map_idx].name, insn_idx);
 999                                 break;
1000                         }
1001                 }
1002
1003                 if (map_idx >= nr_maps) {
1004                         pr_warning("bpf relocation: map_idx %d large than %d\n",
1005                                    (int)map_idx, (int)nr_maps - 1);
1006                         return -LIBBPF_ERRNO__RELOC;
1007                 }
1008
1009                 prog->reloc_desc[i].type = RELO_LD64;
1010                 prog->reloc_desc[i].insn_idx = insn_idx;
1011                 prog->reloc_desc[i].map_idx = map_idx;
1012         }
1013         return 0;
1014 }
1015
1016 static int bpf_map_find_btf_info(struct bpf_map *map, const struct btf *btf)
1017 {
1018         const struct btf_type *container_type;
1019         const struct btf_member *key, *value;
1020         struct bpf_map_def *def = &map->def;
1021         const size_t max_name = 256;
1022         char container_name[max_name];
1023         __s64 key_size, value_size;
1024         __s32 container_id;
1025
1026         if (snprintf(container_name, max_name, "____btf_map_%s", map->name) ==
1027             max_name) {
1028                 pr_warning("map:%s length of '____btf_map_%s' is too long\n",
1029                            map->name, map->name);
1030                 return -EINVAL;
1031         }
1032
1033         container_id = btf__find_by_name(btf, container_name);
1034         if (container_id < 0) {
1035                 pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
1036                          map->name, container_name);
1037                 return container_id;
1038         }
1039
1040         container_type = btf__type_by_id(btf, container_id);
1041         if (!container_type) {
1042                 pr_warning("map:%s cannot find BTF type for container_id:%u\n",
1043                            map->name, container_id);
1044                 return -EINVAL;
1045         }
1046
1047         if (BTF_INFO_KIND(container_type->info) != BTF_KIND_STRUCT ||
1048             BTF_INFO_VLEN(container_type->info) < 2) {
1049                 pr_warning("map:%s container_name:%s is an invalid container struct\n",
1050                            map->name, container_name);
1051                 return -EINVAL;
1052         }
1053
1054         key = (struct btf_member *)(container_type + 1);
1055         value = key + 1;
1056
1057         key_size = btf__resolve_size(btf, key->type);
1058         if (key_size < 0) {
1059                 pr_warning("map:%s invalid BTF key_type_size\n",
1060                            map->name);
1061                 return key_size;
1062         }
1063
1064         if (def->key_size != key_size) {
1065                 pr_warning("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
1066                            map->name, (__u32)key_size, def->key_size);
1067                 return -EINVAL;
1068         }
1069
1070         value_size = btf__resolve_size(btf, value->type);
1071         if (value_size < 0) {
1072                 pr_warning("map:%s invalid BTF value_type_size\n", map->name);
1073                 return value_size;
1074         }
1075
1076         if (def->value_size != value_size) {
1077                 pr_warning("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
1078                            map->name, (__u32)value_size, def->value_size);
1079                 return -EINVAL;
1080         }
1081
1082         map->btf_key_type_id = key->type;
1083         map->btf_value_type_id = value->type;
1084
1085         return 0;
1086 }
1087
1088 static int
1089 bpf_object__create_maps(struct bpf_object *obj)
1090 {
1091         struct bpf_create_map_attr create_attr = {};
1092         unsigned int i;
1093         int err;
1094
1095         for (i = 0; i < obj->nr_maps; i++) {
1096                 struct bpf_map *map = &obj->maps[i];
1097                 struct bpf_map_def *def = &map->def;
1098                 int *pfd = &map->fd;
1099
1100                 create_attr.name = map->name;
1101                 create_attr.map_ifindex = map->map_ifindex;
1102                 create_attr.map_type = def->type;
1103                 create_attr.map_flags = def->map_flags;
1104                 create_attr.key_size = def->key_size;
1105                 create_attr.value_size = def->value_size;
1106                 create_attr.max_entries = def->max_entries;
1107                 create_attr.btf_fd = 0;
1108                 create_attr.btf_key_type_id = 0;
1109                 create_attr.btf_value_type_id = 0;
1110
1111                 if (obj->btf && !bpf_map_find_btf_info(map, obj->btf)) {
1112                         create_attr.btf_fd = btf__fd(obj->btf);
1113                         create_attr.btf_key_type_id = map->btf_key_type_id;
1114                         create_attr.btf_value_type_id = map->btf_value_type_id;
1115                 }
1116
1117                 *pfd = bpf_create_map_xattr(&create_attr);
1118                 if (*pfd < 0 && create_attr.btf_key_type_id) {
1119                         pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
1120                                    map->name, strerror(errno), errno);
1121                         create_attr.btf_fd = 0;
1122                         create_attr.btf_key_type_id = 0;
1123                         create_attr.btf_value_type_id = 0;
1124                         map->btf_key_type_id = 0;
1125                         map->btf_value_type_id = 0;
1126                         *pfd = bpf_create_map_xattr(&create_attr);
1127                 }
1128
1129                 if (*pfd < 0) {
1130                         size_t j;
1131
1132                         err = *pfd;
1133                         pr_warning("failed to create map (name: '%s'): %s\n",
1134                                    map->name,
1135                                    strerror(errno));
1136                         for (j = 0; j < i; j++)
1137                                 zclose(obj->maps[j].fd);
1138                         return err;
1139                 }
1140                 pr_debug("create map %s: fd=%d\n", map->name, *pfd);
1141         }
1142
1143         return 0;
1144 }
1145
1146 static int
1147 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
1148                         struct reloc_desc *relo)
1149 {
1150         struct bpf_insn *insn, *new_insn;
1151         struct bpf_program *text;
1152         size_t new_cnt;
1153
1154         if (relo->type != RELO_CALL)
1155                 return -LIBBPF_ERRNO__RELOC;
1156
1157         if (prog->idx == obj->efile.text_shndx) {
1158                 pr_warning("relo in .text insn %d into off %d\n",
1159                            relo->insn_idx, relo->text_off);
1160                 return -LIBBPF_ERRNO__RELOC;
1161         }
1162
1163         if (prog->main_prog_cnt == 0) {
1164                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
1165                 if (!text) {
1166                         pr_warning("no .text section found yet relo into text exist\n");
1167                         return -LIBBPF_ERRNO__RELOC;
1168                 }
1169                 new_cnt = prog->insns_cnt + text->insns_cnt;
1170                 new_insn = realloc(prog->insns, new_cnt * sizeof(*insn));
1171                 if (!new_insn) {
1172                         pr_warning("oom in prog realloc\n");
1173                         return -ENOMEM;
1174                 }
1175                 memcpy(new_insn + prog->insns_cnt, text->insns,
1176                        text->insns_cnt * sizeof(*insn));
1177                 prog->insns = new_insn;
1178                 prog->main_prog_cnt = prog->insns_cnt;
1179                 prog->insns_cnt = new_cnt;
1180                 pr_debug("added %zd insn from %s to prog %s\n",
1181                          text->insns_cnt, text->section_name,
1182                          prog->section_name);
1183         }
1184         insn = &prog->insns[relo->insn_idx];
1185         insn->imm += prog->main_prog_cnt - relo->insn_idx;
1186         return 0;
1187 }
1188
1189 static int
1190 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
1191 {
1192         int i, err;
1193
1194         if (!prog || !prog->reloc_desc)
1195                 return 0;
1196
1197         for (i = 0; i < prog->nr_reloc; i++) {
1198                 if (prog->reloc_desc[i].type == RELO_LD64) {
1199                         struct bpf_insn *insns = prog->insns;
1200                         int insn_idx, map_idx;
1201
1202                         insn_idx = prog->reloc_desc[i].insn_idx;
1203                         map_idx = prog->reloc_desc[i].map_idx;
1204
1205                         if (insn_idx >= (int)prog->insns_cnt) {
1206                                 pr_warning("relocation out of range: '%s'\n",
1207                                            prog->section_name);
1208                                 return -LIBBPF_ERRNO__RELOC;
1209                         }
1210                         insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
1211                         insns[insn_idx].imm = obj->maps[map_idx].fd;
1212                 } else {
1213                         err = bpf_program__reloc_text(prog, obj,
1214                                                       &prog->reloc_desc[i]);
1215                         if (err)
1216                                 return err;
1217                 }
1218         }
1219
1220         zfree(&prog->reloc_desc);
1221         prog->nr_reloc = 0;
1222         return 0;
1223 }
1224
1225
1226 static int
1227 bpf_object__relocate(struct bpf_object *obj)
1228 {
1229         struct bpf_program *prog;
1230         size_t i;
1231         int err;
1232
1233         for (i = 0; i < obj->nr_programs; i++) {
1234                 prog = &obj->programs[i];
1235
1236                 err = bpf_program__relocate(prog, obj);
1237                 if (err) {
1238                         pr_warning("failed to relocate '%s'\n",
1239                                    prog->section_name);
1240                         return err;
1241                 }
1242         }
1243         return 0;
1244 }
1245
1246 static int bpf_object__collect_reloc(struct bpf_object *obj)
1247 {
1248         int i, err;
1249
1250         if (!obj_elf_valid(obj)) {
1251                 pr_warning("Internal error: elf object is closed\n");
1252                 return -LIBBPF_ERRNO__INTERNAL;
1253         }
1254
1255         for (i = 0; i < obj->efile.nr_reloc; i++) {
1256                 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
1257                 Elf_Data *data = obj->efile.reloc[i].data;
1258                 int idx = shdr->sh_info;
1259                 struct bpf_program *prog;
1260
1261                 if (shdr->sh_type != SHT_REL) {
1262                         pr_warning("internal error at %d\n", __LINE__);
1263                         return -LIBBPF_ERRNO__INTERNAL;
1264                 }
1265
1266                 prog = bpf_object__find_prog_by_idx(obj, idx);
1267                 if (!prog) {
1268                         pr_warning("relocation failed: no section(%d)\n", idx);
1269                         return -LIBBPF_ERRNO__RELOC;
1270                 }
1271
1272                 err = bpf_program__collect_reloc(prog,
1273                                                  shdr, data,
1274                                                  obj);
1275                 if (err)
1276                         return err;
1277         }
1278         return 0;
1279 }
1280
1281 static int
1282 load_program(enum bpf_prog_type type, enum bpf_attach_type expected_attach_type,
1283              const char *name, struct bpf_insn *insns, int insns_cnt,
1284              char *license, u32 kern_version, int *pfd, int prog_ifindex)
1285 {
1286         struct bpf_load_program_attr load_attr;
1287         char *log_buf;
1288         int ret;
1289
1290         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
1291         load_attr.prog_type = type;
1292         load_attr.expected_attach_type = expected_attach_type;
1293         load_attr.name = name;
1294         load_attr.insns = insns;
1295         load_attr.insns_cnt = insns_cnt;
1296         load_attr.license = license;
1297         load_attr.kern_version = kern_version;
1298         load_attr.prog_ifindex = prog_ifindex;
1299
1300         if (!load_attr.insns || !load_attr.insns_cnt)
1301                 return -EINVAL;
1302
1303         log_buf = malloc(BPF_LOG_BUF_SIZE);
1304         if (!log_buf)
1305                 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
1306
1307         ret = bpf_load_program_xattr(&load_attr, log_buf, BPF_LOG_BUF_SIZE);
1308
1309         if (ret >= 0) {
1310                 *pfd = ret;
1311                 ret = 0;
1312                 goto out;
1313         }
1314
1315         ret = -LIBBPF_ERRNO__LOAD;
1316         pr_warning("load bpf program failed: %s\n", strerror(errno));
1317
1318         if (log_buf && log_buf[0] != '\0') {
1319                 ret = -LIBBPF_ERRNO__VERIFY;
1320                 pr_warning("-- BEGIN DUMP LOG ---\n");
1321                 pr_warning("\n%s\n", log_buf);
1322                 pr_warning("-- END LOG --\n");
1323         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
1324                 pr_warning("Program too large (%zu insns), at most %d insns\n",
1325                            load_attr.insns_cnt, BPF_MAXINSNS);
1326                 ret = -LIBBPF_ERRNO__PROG2BIG;
1327         } else {
1328                 /* Wrong program type? */
1329                 if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
1330                         int fd;
1331
1332                         load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
1333                         load_attr.expected_attach_type = 0;
1334                         fd = bpf_load_program_xattr(&load_attr, NULL, 0);
1335                         if (fd >= 0) {
1336                                 close(fd);
1337                                 ret = -LIBBPF_ERRNO__PROGTYPE;
1338                                 goto out;
1339                         }
1340                 }
1341
1342                 if (log_buf)
1343                         ret = -LIBBPF_ERRNO__KVER;
1344         }
1345
1346 out:
1347         free(log_buf);
1348         return ret;
1349 }
1350
1351 static int
1352 bpf_program__load(struct bpf_program *prog,
1353                   char *license, u32 kern_version)
1354 {
1355         int err = 0, fd, i;
1356
1357         if (prog->instances.nr < 0 || !prog->instances.fds) {
1358                 if (prog->preprocessor) {
1359                         pr_warning("Internal error: can't load program '%s'\n",
1360                                    prog->section_name);
1361                         return -LIBBPF_ERRNO__INTERNAL;
1362                 }
1363
1364                 prog->instances.fds = malloc(sizeof(int));
1365                 if (!prog->instances.fds) {
1366                         pr_warning("Not enough memory for BPF fds\n");
1367                         return -ENOMEM;
1368                 }
1369                 prog->instances.nr = 1;
1370                 prog->instances.fds[0] = -1;
1371         }
1372
1373         if (!prog->preprocessor) {
1374                 if (prog->instances.nr != 1) {
1375                         pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
1376                                    prog->section_name, prog->instances.nr);
1377                 }
1378                 err = load_program(prog->type, prog->expected_attach_type,
1379                                    prog->name, prog->insns, prog->insns_cnt,
1380                                    license, kern_version, &fd,
1381                                    prog->prog_ifindex);
1382                 if (!err)
1383                         prog->instances.fds[0] = fd;
1384                 goto out;
1385         }
1386
1387         for (i = 0; i < prog->instances.nr; i++) {
1388                 struct bpf_prog_prep_result result;
1389                 bpf_program_prep_t preprocessor = prog->preprocessor;
1390
1391                 bzero(&result, sizeof(result));
1392                 err = preprocessor(prog, i, prog->insns,
1393                                    prog->insns_cnt, &result);
1394                 if (err) {
1395                         pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
1396                                    i, prog->section_name);
1397                         goto out;
1398                 }
1399
1400                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
1401                         pr_debug("Skip loading the %dth instance of program '%s'\n",
1402                                  i, prog->section_name);
1403                         prog->instances.fds[i] = -1;
1404                         if (result.pfd)
1405                                 *result.pfd = -1;
1406                         continue;
1407                 }
1408
1409                 err = load_program(prog->type, prog->expected_attach_type,
1410                                    prog->name, result.new_insn_ptr,
1411                                    result.new_insn_cnt,
1412                                    license, kern_version, &fd,
1413                                    prog->prog_ifindex);
1414
1415                 if (err) {
1416                         pr_warning("Loading the %dth instance of program '%s' failed\n",
1417                                         i, prog->section_name);
1418                         goto out;
1419                 }
1420
1421                 if (result.pfd)
1422                         *result.pfd = fd;
1423                 prog->instances.fds[i] = fd;
1424         }
1425 out:
1426         if (err)
1427                 pr_warning("failed to load program '%s'\n",
1428                            prog->section_name);
1429         zfree(&prog->insns);
1430         prog->insns_cnt = 0;
1431         return err;
1432 }
1433
1434 static int
1435 bpf_object__load_progs(struct bpf_object *obj)
1436 {
1437         size_t i;
1438         int err;
1439
1440         for (i = 0; i < obj->nr_programs; i++) {
1441                 if (obj->programs[i].idx == obj->efile.text_shndx)
1442                         continue;
1443                 err = bpf_program__load(&obj->programs[i],
1444                                         obj->license,
1445                                         obj->kern_version);
1446                 if (err)
1447                         return err;
1448         }
1449         return 0;
1450 }
1451
1452 static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
1453 {
1454         switch (type) {
1455         case BPF_PROG_TYPE_SOCKET_FILTER:
1456         case BPF_PROG_TYPE_SCHED_CLS:
1457         case BPF_PROG_TYPE_SCHED_ACT:
1458         case BPF_PROG_TYPE_XDP:
1459         case BPF_PROG_TYPE_CGROUP_SKB:
1460         case BPF_PROG_TYPE_CGROUP_SOCK:
1461         case BPF_PROG_TYPE_LWT_IN:
1462         case BPF_PROG_TYPE_LWT_OUT:
1463         case BPF_PROG_TYPE_LWT_XMIT:
1464         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
1465         case BPF_PROG_TYPE_SOCK_OPS:
1466         case BPF_PROG_TYPE_SK_SKB:
1467         case BPF_PROG_TYPE_CGROUP_DEVICE:
1468         case BPF_PROG_TYPE_SK_MSG:
1469         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1470         case BPF_PROG_TYPE_LIRC_MODE2:
1471                 return false;
1472         case BPF_PROG_TYPE_UNSPEC:
1473         case BPF_PROG_TYPE_KPROBE:
1474         case BPF_PROG_TYPE_TRACEPOINT:
1475         case BPF_PROG_TYPE_PERF_EVENT:
1476         case BPF_PROG_TYPE_RAW_TRACEPOINT:
1477         default:
1478                 return true;
1479         }
1480 }
1481
1482 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
1483 {
1484         if (needs_kver && obj->kern_version == 0) {
1485                 pr_warning("%s doesn't provide kernel version\n",
1486                            obj->path);
1487                 return -LIBBPF_ERRNO__KVERSION;
1488         }
1489         return 0;
1490 }
1491
1492 static struct bpf_object *
1493 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
1494                    bool needs_kver)
1495 {
1496         struct bpf_object *obj;
1497         int err;
1498
1499         if (elf_version(EV_CURRENT) == EV_NONE) {
1500                 pr_warning("failed to init libelf for %s\n", path);
1501                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1502         }
1503
1504         obj = bpf_object__new(path, obj_buf, obj_buf_sz);
1505         if (IS_ERR(obj))
1506                 return obj;
1507
1508         CHECK_ERR(bpf_object__elf_init(obj), err, out);
1509         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
1510         CHECK_ERR(bpf_object__elf_collect(obj), err, out);
1511         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
1512         CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
1513
1514         bpf_object__elf_finish(obj);
1515         return obj;
1516 out:
1517         bpf_object__close(obj);
1518         return ERR_PTR(err);
1519 }
1520
1521 struct bpf_object *bpf_object__open(const char *path)
1522 {
1523         /* param validation */
1524         if (!path)
1525                 return NULL;
1526
1527         pr_debug("loading %s\n", path);
1528
1529         return __bpf_object__open(path, NULL, 0, true);
1530 }
1531
1532 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
1533                                            size_t obj_buf_sz,
1534                                            const char *name)
1535 {
1536         char tmp_name[64];
1537
1538         /* param validation */
1539         if (!obj_buf || obj_buf_sz <= 0)
1540                 return NULL;
1541
1542         if (!name) {
1543                 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
1544                          (unsigned long)obj_buf,
1545                          (unsigned long)obj_buf_sz);
1546                 tmp_name[sizeof(tmp_name) - 1] = '\0';
1547                 name = tmp_name;
1548         }
1549         pr_debug("loading object '%s' from buffer\n",
1550                  name);
1551
1552         return __bpf_object__open(name, obj_buf, obj_buf_sz, true);
1553 }
1554
1555 int bpf_object__unload(struct bpf_object *obj)
1556 {
1557         size_t i;
1558
1559         if (!obj)
1560                 return -EINVAL;
1561
1562         for (i = 0; i < obj->nr_maps; i++)
1563                 zclose(obj->maps[i].fd);
1564
1565         for (i = 0; i < obj->nr_programs; i++)
1566                 bpf_program__unload(&obj->programs[i]);
1567
1568         return 0;
1569 }
1570
1571 int bpf_object__load(struct bpf_object *obj)
1572 {
1573         int err;
1574
1575         if (!obj)
1576                 return -EINVAL;
1577
1578         if (obj->loaded) {
1579                 pr_warning("object should not be loaded twice\n");
1580                 return -EINVAL;
1581         }
1582
1583         obj->loaded = true;
1584
1585         CHECK_ERR(bpf_object__create_maps(obj), err, out);
1586         CHECK_ERR(bpf_object__relocate(obj), err, out);
1587         CHECK_ERR(bpf_object__load_progs(obj), err, out);
1588
1589         return 0;
1590 out:
1591         bpf_object__unload(obj);
1592         pr_warning("failed to load object '%s'\n", obj->path);
1593         return err;
1594 }
1595
1596 static int check_path(const char *path)
1597 {
1598         struct statfs st_fs;
1599         char *dname, *dir;
1600         int err = 0;
1601
1602         if (path == NULL)
1603                 return -EINVAL;
1604
1605         dname = strdup(path);
1606         if (dname == NULL)
1607                 return -ENOMEM;
1608
1609         dir = dirname(dname);
1610         if (statfs(dir, &st_fs)) {
1611                 pr_warning("failed to statfs %s: %s\n", dir, strerror(errno));
1612                 err = -errno;
1613         }
1614         free(dname);
1615
1616         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
1617                 pr_warning("specified path %s is not on BPF FS\n", path);
1618                 err = -EINVAL;
1619         }
1620
1621         return err;
1622 }
1623
1624 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
1625                               int instance)
1626 {
1627         int err;
1628
1629         err = check_path(path);
1630         if (err)
1631                 return err;
1632
1633         if (prog == NULL) {
1634                 pr_warning("invalid program pointer\n");
1635                 return -EINVAL;
1636         }
1637
1638         if (instance < 0 || instance >= prog->instances.nr) {
1639                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
1640                            instance, prog->section_name, prog->instances.nr);
1641                 return -EINVAL;
1642         }
1643
1644         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
1645                 pr_warning("failed to pin program: %s\n", strerror(errno));
1646                 return -errno;
1647         }
1648         pr_debug("pinned program '%s'\n", path);
1649
1650         return 0;
1651 }
1652
1653 static int make_dir(const char *path)
1654 {
1655         int err = 0;
1656
1657         if (mkdir(path, 0700) && errno != EEXIST)
1658                 err = -errno;
1659
1660         if (err)
1661                 pr_warning("failed to mkdir %s: %s\n", path, strerror(-err));
1662         return err;
1663 }
1664
1665 int bpf_program__pin(struct bpf_program *prog, const char *path)
1666 {
1667         int i, err;
1668
1669         err = check_path(path);
1670         if (err)
1671                 return err;
1672
1673         if (prog == NULL) {
1674                 pr_warning("invalid program pointer\n");
1675                 return -EINVAL;
1676         }
1677
1678         if (prog->instances.nr <= 0) {
1679                 pr_warning("no instances of prog %s to pin\n",
1680                            prog->section_name);
1681                 return -EINVAL;
1682         }
1683
1684         err = make_dir(path);
1685         if (err)
1686                 return err;
1687
1688         for (i = 0; i < prog->instances.nr; i++) {
1689                 char buf[PATH_MAX];
1690                 int len;
1691
1692                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
1693                 if (len < 0)
1694                         return -EINVAL;
1695                 else if (len >= PATH_MAX)
1696                         return -ENAMETOOLONG;
1697
1698                 err = bpf_program__pin_instance(prog, buf, i);
1699                 if (err)
1700                         return err;
1701         }
1702
1703         return 0;
1704 }
1705
1706 int bpf_map__pin(struct bpf_map *map, const char *path)
1707 {
1708         int err;
1709
1710         err = check_path(path);
1711         if (err)
1712                 return err;
1713
1714         if (map == NULL) {
1715                 pr_warning("invalid map pointer\n");
1716                 return -EINVAL;
1717         }
1718
1719         if (bpf_obj_pin(map->fd, path)) {
1720                 pr_warning("failed to pin map: %s\n", strerror(errno));
1721                 return -errno;
1722         }
1723
1724         pr_debug("pinned map '%s'\n", path);
1725         return 0;
1726 }
1727
1728 int bpf_object__pin(struct bpf_object *obj, const char *path)
1729 {
1730         struct bpf_program *prog;
1731         struct bpf_map *map;
1732         int err;
1733
1734         if (!obj)
1735                 return -ENOENT;
1736
1737         if (!obj->loaded) {
1738                 pr_warning("object not yet loaded; load it first\n");
1739                 return -ENOENT;
1740         }
1741
1742         err = make_dir(path);
1743         if (err)
1744                 return err;
1745
1746         bpf_map__for_each(map, obj) {
1747                 char buf[PATH_MAX];
1748                 int len;
1749
1750                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1751                                bpf_map__name(map));
1752                 if (len < 0)
1753                         return -EINVAL;
1754                 else if (len >= PATH_MAX)
1755                         return -ENAMETOOLONG;
1756
1757                 err = bpf_map__pin(map, buf);
1758                 if (err)
1759                         return err;
1760         }
1761
1762         bpf_object__for_each_program(prog, obj) {
1763                 char buf[PATH_MAX];
1764                 int len;
1765
1766                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1767                                prog->section_name);
1768                 if (len < 0)
1769                         return -EINVAL;
1770                 else if (len >= PATH_MAX)
1771                         return -ENAMETOOLONG;
1772
1773                 err = bpf_program__pin(prog, buf);
1774                 if (err)
1775                         return err;
1776         }
1777
1778         return 0;
1779 }
1780
1781 void bpf_object__close(struct bpf_object *obj)
1782 {
1783         size_t i;
1784
1785         if (!obj)
1786                 return;
1787
1788         if (obj->clear_priv)
1789                 obj->clear_priv(obj, obj->priv);
1790
1791         bpf_object__elf_finish(obj);
1792         bpf_object__unload(obj);
1793         btf__free(obj->btf);
1794
1795         for (i = 0; i < obj->nr_maps; i++) {
1796                 zfree(&obj->maps[i].name);
1797                 if (obj->maps[i].clear_priv)
1798                         obj->maps[i].clear_priv(&obj->maps[i],
1799                                                 obj->maps[i].priv);
1800                 obj->maps[i].priv = NULL;
1801                 obj->maps[i].clear_priv = NULL;
1802         }
1803         zfree(&obj->maps);
1804         obj->nr_maps = 0;
1805
1806         if (obj->programs && obj->nr_programs) {
1807                 for (i = 0; i < obj->nr_programs; i++)
1808                         bpf_program__exit(&obj->programs[i]);
1809         }
1810         zfree(&obj->programs);
1811
1812         list_del(&obj->list);
1813         free(obj);
1814 }
1815
1816 struct bpf_object *
1817 bpf_object__next(struct bpf_object *prev)
1818 {
1819         struct bpf_object *next;
1820
1821         if (!prev)
1822                 next = list_first_entry(&bpf_objects_list,
1823                                         struct bpf_object,
1824                                         list);
1825         else
1826                 next = list_next_entry(prev, list);
1827
1828         /* Empty list is noticed here so don't need checking on entry. */
1829         if (&next->list == &bpf_objects_list)
1830                 return NULL;
1831
1832         return next;
1833 }
1834
1835 const char *bpf_object__name(struct bpf_object *obj)
1836 {
1837         return obj ? obj->path : ERR_PTR(-EINVAL);
1838 }
1839
1840 unsigned int bpf_object__kversion(struct bpf_object *obj)
1841 {
1842         return obj ? obj->kern_version : 0;
1843 }
1844
1845 int bpf_object__btf_fd(const struct bpf_object *obj)
1846 {
1847         return obj->btf ? btf__fd(obj->btf) : -1;
1848 }
1849
1850 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
1851                          bpf_object_clear_priv_t clear_priv)
1852 {
1853         if (obj->priv && obj->clear_priv)
1854                 obj->clear_priv(obj, obj->priv);
1855
1856         obj->priv = priv;
1857         obj->clear_priv = clear_priv;
1858         return 0;
1859 }
1860
1861 void *bpf_object__priv(struct bpf_object *obj)
1862 {
1863         return obj ? obj->priv : ERR_PTR(-EINVAL);
1864 }
1865
1866 struct bpf_program *
1867 bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1868 {
1869         size_t idx;
1870
1871         if (!obj->programs)
1872                 return NULL;
1873         /* First handler */
1874         if (prev == NULL)
1875                 return &obj->programs[0];
1876
1877         if (prev->obj != obj) {
1878                 pr_warning("error: program handler doesn't match object\n");
1879                 return NULL;
1880         }
1881
1882         idx = (prev - obj->programs) + 1;
1883         if (idx >= obj->nr_programs)
1884                 return NULL;
1885         return &obj->programs[idx];
1886 }
1887
1888 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
1889                           bpf_program_clear_priv_t clear_priv)
1890 {
1891         if (prog->priv && prog->clear_priv)
1892                 prog->clear_priv(prog, prog->priv);
1893
1894         prog->priv = priv;
1895         prog->clear_priv = clear_priv;
1896         return 0;
1897 }
1898
1899 void *bpf_program__priv(struct bpf_program *prog)
1900 {
1901         return prog ? prog->priv : ERR_PTR(-EINVAL);
1902 }
1903
1904 const char *bpf_program__title(struct bpf_program *prog, bool needs_copy)
1905 {
1906         const char *title;
1907
1908         title = prog->section_name;
1909         if (needs_copy) {
1910                 title = strdup(title);
1911                 if (!title) {
1912                         pr_warning("failed to strdup program title\n");
1913                         return ERR_PTR(-ENOMEM);
1914                 }
1915         }
1916
1917         return title;
1918 }
1919
1920 int bpf_program__fd(struct bpf_program *prog)
1921 {
1922         return bpf_program__nth_fd(prog, 0);
1923 }
1924
1925 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
1926                           bpf_program_prep_t prep)
1927 {
1928         int *instances_fds;
1929
1930         if (nr_instances <= 0 || !prep)
1931                 return -EINVAL;
1932
1933         if (prog->instances.nr > 0 || prog->instances.fds) {
1934                 pr_warning("Can't set pre-processor after loading\n");
1935                 return -EINVAL;
1936         }
1937
1938         instances_fds = malloc(sizeof(int) * nr_instances);
1939         if (!instances_fds) {
1940                 pr_warning("alloc memory failed for fds\n");
1941                 return -ENOMEM;
1942         }
1943
1944         /* fill all fd with -1 */
1945         memset(instances_fds, -1, sizeof(int) * nr_instances);
1946
1947         prog->instances.nr = nr_instances;
1948         prog->instances.fds = instances_fds;
1949         prog->preprocessor = prep;
1950         return 0;
1951 }
1952
1953 int bpf_program__nth_fd(struct bpf_program *prog, int n)
1954 {
1955         int fd;
1956
1957         if (n >= prog->instances.nr || n < 0) {
1958                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
1959                            n, prog->section_name, prog->instances.nr);
1960                 return -EINVAL;
1961         }
1962
1963         fd = prog->instances.fds[n];
1964         if (fd < 0) {
1965                 pr_warning("%dth instance of program '%s' is invalid\n",
1966                            n, prog->section_name);
1967                 return -ENOENT;
1968         }
1969
1970         return fd;
1971 }
1972
1973 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
1974 {
1975         prog->type = type;
1976 }
1977
1978 static bool bpf_program__is_type(struct bpf_program *prog,
1979                                  enum bpf_prog_type type)
1980 {
1981         return prog ? (prog->type == type) : false;
1982 }
1983
1984 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                   \
1985 int bpf_program__set_##NAME(struct bpf_program *prog)   \
1986 {                                                       \
1987         if (!prog)                                      \
1988                 return -EINVAL;                         \
1989         bpf_program__set_type(prog, TYPE);              \
1990         return 0;                                       \
1991 }                                                       \
1992                                                         \
1993 bool bpf_program__is_##NAME(struct bpf_program *prog)   \
1994 {                                                       \
1995         return bpf_program__is_type(prog, TYPE);        \
1996 }                                                       \
1997
1998 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
1999 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
2000 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
2001 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
2002 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
2003 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
2004 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
2005 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
2006
2007 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
2008                                            enum bpf_attach_type type)
2009 {
2010         prog->expected_attach_type = type;
2011 }
2012
2013 #define BPF_PROG_SEC_FULL(string, ptype, atype) \
2014         { string, sizeof(string) - 1, ptype, atype }
2015
2016 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_FULL(string, ptype, 0)
2017
2018 #define BPF_S_PROG_SEC(string, ptype) \
2019         BPF_PROG_SEC_FULL(string, BPF_PROG_TYPE_CGROUP_SOCK, ptype)
2020
2021 #define BPF_SA_PROG_SEC(string, ptype) \
2022         BPF_PROG_SEC_FULL(string, BPF_PROG_TYPE_CGROUP_SOCK_ADDR, ptype)
2023
2024 static const struct {
2025         const char *sec;
2026         size_t len;
2027         enum bpf_prog_type prog_type;
2028         enum bpf_attach_type expected_attach_type;
2029 } section_names[] = {
2030         BPF_PROG_SEC("socket",          BPF_PROG_TYPE_SOCKET_FILTER),
2031         BPF_PROG_SEC("kprobe/",         BPF_PROG_TYPE_KPROBE),
2032         BPF_PROG_SEC("kretprobe/",      BPF_PROG_TYPE_KPROBE),
2033         BPF_PROG_SEC("classifier",      BPF_PROG_TYPE_SCHED_CLS),
2034         BPF_PROG_SEC("action",          BPF_PROG_TYPE_SCHED_ACT),
2035         BPF_PROG_SEC("tracepoint/",     BPF_PROG_TYPE_TRACEPOINT),
2036         BPF_PROG_SEC("raw_tracepoint/", BPF_PROG_TYPE_RAW_TRACEPOINT),
2037         BPF_PROG_SEC("xdp",             BPF_PROG_TYPE_XDP),
2038         BPF_PROG_SEC("perf_event",      BPF_PROG_TYPE_PERF_EVENT),
2039         BPF_PROG_SEC("cgroup/skb",      BPF_PROG_TYPE_CGROUP_SKB),
2040         BPF_PROG_SEC("cgroup/sock",     BPF_PROG_TYPE_CGROUP_SOCK),
2041         BPF_PROG_SEC("cgroup/dev",      BPF_PROG_TYPE_CGROUP_DEVICE),
2042         BPF_PROG_SEC("lwt_in",          BPF_PROG_TYPE_LWT_IN),
2043         BPF_PROG_SEC("lwt_out",         BPF_PROG_TYPE_LWT_OUT),
2044         BPF_PROG_SEC("lwt_xmit",        BPF_PROG_TYPE_LWT_XMIT),
2045         BPF_PROG_SEC("sockops",         BPF_PROG_TYPE_SOCK_OPS),
2046         BPF_PROG_SEC("sk_skb",          BPF_PROG_TYPE_SK_SKB),
2047         BPF_PROG_SEC("sk_msg",          BPF_PROG_TYPE_SK_MSG),
2048         BPF_SA_PROG_SEC("cgroup/bind4", BPF_CGROUP_INET4_BIND),
2049         BPF_SA_PROG_SEC("cgroup/bind6", BPF_CGROUP_INET6_BIND),
2050         BPF_SA_PROG_SEC("cgroup/connect4", BPF_CGROUP_INET4_CONNECT),
2051         BPF_SA_PROG_SEC("cgroup/connect6", BPF_CGROUP_INET6_CONNECT),
2052         BPF_SA_PROG_SEC("cgroup/sendmsg4", BPF_CGROUP_UDP4_SENDMSG),
2053         BPF_SA_PROG_SEC("cgroup/sendmsg6", BPF_CGROUP_UDP6_SENDMSG),
2054         BPF_S_PROG_SEC("cgroup/post_bind4", BPF_CGROUP_INET4_POST_BIND),
2055         BPF_S_PROG_SEC("cgroup/post_bind6", BPF_CGROUP_INET6_POST_BIND),
2056 };
2057
2058 #undef BPF_PROG_SEC
2059 #undef BPF_PROG_SEC_FULL
2060 #undef BPF_S_PROG_SEC
2061 #undef BPF_SA_PROG_SEC
2062
2063 static int bpf_program__identify_section(struct bpf_program *prog)
2064 {
2065         int i;
2066
2067         if (!prog->section_name)
2068                 goto err;
2069
2070         for (i = 0; i < ARRAY_SIZE(section_names); i++)
2071                 if (strncmp(prog->section_name, section_names[i].sec,
2072                             section_names[i].len) == 0)
2073                         return i;
2074
2075 err:
2076         pr_warning("failed to guess program type based on section name %s\n",
2077                    prog->section_name);
2078
2079         return -1;
2080 }
2081
2082 int bpf_map__fd(struct bpf_map *map)
2083 {
2084         return map ? map->fd : -EINVAL;
2085 }
2086
2087 const struct bpf_map_def *bpf_map__def(struct bpf_map *map)
2088 {
2089         return map ? &map->def : ERR_PTR(-EINVAL);
2090 }
2091
2092 const char *bpf_map__name(struct bpf_map *map)
2093 {
2094         return map ? map->name : NULL;
2095 }
2096
2097 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
2098 {
2099         return map ? map->btf_key_type_id : 0;
2100 }
2101
2102 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
2103 {
2104         return map ? map->btf_value_type_id : 0;
2105 }
2106
2107 int bpf_map__set_priv(struct bpf_map *map, void *priv,
2108                      bpf_map_clear_priv_t clear_priv)
2109 {
2110         if (!map)
2111                 return -EINVAL;
2112
2113         if (map->priv) {
2114                 if (map->clear_priv)
2115                         map->clear_priv(map, map->priv);
2116         }
2117
2118         map->priv = priv;
2119         map->clear_priv = clear_priv;
2120         return 0;
2121 }
2122
2123 void *bpf_map__priv(struct bpf_map *map)
2124 {
2125         return map ? map->priv : ERR_PTR(-EINVAL);
2126 }
2127
2128 struct bpf_map *
2129 bpf_map__next(struct bpf_map *prev, struct bpf_object *obj)
2130 {
2131         size_t idx;
2132         struct bpf_map *s, *e;
2133
2134         if (!obj || !obj->maps)
2135                 return NULL;
2136
2137         s = obj->maps;
2138         e = obj->maps + obj->nr_maps;
2139
2140         if (prev == NULL)
2141                 return s;
2142
2143         if ((prev < s) || (prev >= e)) {
2144                 pr_warning("error in %s: map handler doesn't belong to object\n",
2145                            __func__);
2146                 return NULL;
2147         }
2148
2149         idx = (prev - obj->maps) + 1;
2150         if (idx >= obj->nr_maps)
2151                 return NULL;
2152         return &obj->maps[idx];
2153 }
2154
2155 struct bpf_map *
2156 bpf_object__find_map_by_name(struct bpf_object *obj, const char *name)
2157 {
2158         struct bpf_map *pos;
2159
2160         bpf_map__for_each(pos, obj) {
2161                 if (pos->name && !strcmp(pos->name, name))
2162                         return pos;
2163         }
2164         return NULL;
2165 }
2166
2167 struct bpf_map *
2168 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
2169 {
2170         int i;
2171
2172         for (i = 0; i < obj->nr_maps; i++) {
2173                 if (obj->maps[i].offset == offset)
2174                         return &obj->maps[i];
2175         }
2176         return ERR_PTR(-ENOENT);
2177 }
2178
2179 long libbpf_get_error(const void *ptr)
2180 {
2181         if (IS_ERR(ptr))
2182                 return PTR_ERR(ptr);
2183         return 0;
2184 }
2185
2186 int bpf_prog_load(const char *file, enum bpf_prog_type type,
2187                   struct bpf_object **pobj, int *prog_fd)
2188 {
2189         struct bpf_prog_load_attr attr;
2190
2191         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
2192         attr.file = file;
2193         attr.prog_type = type;
2194         attr.expected_attach_type = 0;
2195
2196         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
2197 }
2198
2199 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
2200                         struct bpf_object **pobj, int *prog_fd)
2201 {
2202         struct bpf_program *prog, *first_prog = NULL;
2203         enum bpf_attach_type expected_attach_type;
2204         enum bpf_prog_type prog_type;
2205         struct bpf_object *obj;
2206         struct bpf_map *map;
2207         int section_idx;
2208         int err;
2209
2210         if (!attr)
2211                 return -EINVAL;
2212         if (!attr->file)
2213                 return -EINVAL;
2214
2215         obj = __bpf_object__open(attr->file, NULL, 0,
2216                                  bpf_prog_type__needs_kver(attr->prog_type));
2217         if (IS_ERR_OR_NULL(obj))
2218                 return -ENOENT;
2219
2220         bpf_object__for_each_program(prog, obj) {
2221                 /*
2222                  * If type is not specified, try to guess it based on
2223                  * section name.
2224                  */
2225                 prog_type = attr->prog_type;
2226                 prog->prog_ifindex = attr->ifindex;
2227                 expected_attach_type = attr->expected_attach_type;
2228                 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
2229                         section_idx = bpf_program__identify_section(prog);
2230                         if (section_idx < 0) {
2231                                 bpf_object__close(obj);
2232                                 return -EINVAL;
2233                         }
2234                         prog_type = section_names[section_idx].prog_type;
2235                         expected_attach_type =
2236                                 section_names[section_idx].expected_attach_type;
2237                 }
2238
2239                 bpf_program__set_type(prog, prog_type);
2240                 bpf_program__set_expected_attach_type(prog,
2241                                                       expected_attach_type);
2242
2243                 if (prog->idx != obj->efile.text_shndx && !first_prog)
2244                         first_prog = prog;
2245         }
2246
2247         bpf_map__for_each(map, obj) {
2248                 map->map_ifindex = attr->ifindex;
2249         }
2250
2251         if (!first_prog) {
2252                 pr_warning("object file doesn't contain bpf program\n");
2253                 bpf_object__close(obj);
2254                 return -ENOENT;
2255         }
2256
2257         err = bpf_object__load(obj);
2258         if (err) {
2259                 bpf_object__close(obj);
2260                 return -EINVAL;
2261         }
2262
2263         *pobj = obj;
2264         *prog_fd = bpf_program__fd(first_prog);
2265         return 0;
2266 }
2267
2268 enum bpf_perf_event_ret
2269 bpf_perf_event_read_simple(void *mem, unsigned long size,
2270                            unsigned long page_size, void **buf, size_t *buf_len,
2271                            bpf_perf_event_print_t fn, void *priv)
2272 {
2273         volatile struct perf_event_mmap_page *header = mem;
2274         __u64 data_tail = header->data_tail;
2275         __u64 data_head = header->data_head;
2276         int ret = LIBBPF_PERF_EVENT_ERROR;
2277         void *base, *begin, *end;
2278
2279         asm volatile("" ::: "memory"); /* in real code it should be smp_rmb() */
2280         if (data_head == data_tail)
2281                 return LIBBPF_PERF_EVENT_CONT;
2282
2283         base = ((char *)header) + page_size;
2284
2285         begin = base + data_tail % size;
2286         end = base + data_head % size;
2287
2288         while (begin != end) {
2289                 struct perf_event_header *ehdr;
2290
2291                 ehdr = begin;
2292                 if (begin + ehdr->size > base + size) {
2293                         long len = base + size - begin;
2294
2295                         if (*buf_len < ehdr->size) {
2296                                 free(*buf);
2297                                 *buf = malloc(ehdr->size);
2298                                 if (!*buf) {
2299                                         ret = LIBBPF_PERF_EVENT_ERROR;
2300                                         break;
2301                                 }
2302                                 *buf_len = ehdr->size;
2303                         }
2304
2305                         memcpy(*buf, begin, len);
2306                         memcpy(*buf + len, base, ehdr->size - len);
2307                         ehdr = (void *)*buf;
2308                         begin = base + ehdr->size - len;
2309                 } else if (begin + ehdr->size == base + size) {
2310                         begin = base;
2311                 } else {
2312                         begin += ehdr->size;
2313                 }
2314
2315                 ret = fn(ehdr, priv);
2316                 if (ret != LIBBPF_PERF_EVENT_CONT)
2317                         break;
2318
2319                 data_tail += ehdr->size;
2320         }
2321
2322         __sync_synchronize(); /* smp_mb() */
2323         header->data_tail = data_tail;
2324
2325         return ret;
2326 }