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/list.h>
  40 #include <linux/limits.h>
  41 #include <sys/stat.h>
  42 #include <sys/types.h>
  43 #include <sys/vfs.h>
  44 #include <libelf.h>
  45 #include <gelf.h>
  46
  47 #include "libbpf.h"
  48 #include "bpf.h"
  49 #include "btf.h"
  50
  51 #ifndef EM_BPF
  52 #define EM_BPF 247
  53 #endif
  54
  55 #ifndef BPF_FS_MAGIC
  56 #define BPF_FS_MAGIC            0xcafe4a11
  57 #endif
  58
  59 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  60
  61 __printf(1, 2)
  62 static int __base_pr(const char *format, ...)
  63 {
  64         va_list args;
  65         int err;
  66
  67         va_start(args, format);
  68         err = vfprintf(stderr, format, args);
  69         va_end(args);
  70         return err;
  71 }
  72
  73 static __printf(1, 2) libbpf_print_fn_t __pr_warning = __base_pr;
  74 static __printf(1, 2) libbpf_print_fn_t __pr_info = __base_pr;
  75 static __printf(1, 2) libbpf_print_fn_t __pr_debug;
  76
  77 #define __pr(func, fmt, ...)    \
  78 do {                            \
  79         if ((func))             \
  80                 (func)("libbpf: " fmt, ##__VA_ARGS__); \
  81 } while (0)
  82
  83 #define pr_warning(fmt, ...)    __pr(__pr_warning, fmt, ##__VA_ARGS__)
  84 #define pr_info(fmt, ...)       __pr(__pr_info, fmt, ##__VA_ARGS__)
  85 #define pr_debug(fmt, ...)      __pr(__pr_debug, fmt, ##__VA_ARGS__)
  86
  87 void libbpf_set_print(libbpf_print_fn_t warn,
  88                       libbpf_print_fn_t info,
  89                       libbpf_print_fn_t debug)
  90 {
  91         __pr_warning = warn;
  92         __pr_info = info;
  93         __pr_debug = debug;
  94 }
  95
  96 #define STRERR_BUFSIZE  128
  97
  98 #define ERRNO_OFFSET(e)         ((e) - __LIBBPF_ERRNO__START)
  99 #define ERRCODE_OFFSET(c)       ERRNO_OFFSET(LIBBPF_ERRNO__##c)
 100 #define NR_ERRNO        (__LIBBPF_ERRNO__END - __LIBBPF_ERRNO__START)
 101
 102 static const char *libbpf_strerror_table[NR_ERRNO] = {
 103         [ERRCODE_OFFSET(LIBELF)]        = "Something wrong in libelf",
 104         [ERRCODE_OFFSET(FORMAT)]        = "BPF object format invalid",
 105         [ERRCODE_OFFSET(KVERSION)]      = "'version' section incorrect or lost",
 106         [ERRCODE_OFFSET(ENDIAN)]        = "Endian mismatch",
 107         [ERRCODE_OFFSET(INTERNAL)]      = "Internal error in libbpf",
 108         [ERRCODE_OFFSET(RELOC)]         = "Relocation failed",
 109         [ERRCODE_OFFSET(VERIFY)]        = "Kernel verifier blocks program loading",
 110         [ERRCODE_OFFSET(PROG2BIG)]      = "Program too big",
 111         [ERRCODE_OFFSET(KVER)]          = "Incorrect kernel version",
 112         [ERRCODE_OFFSET(PROGTYPE)]      = "Kernel doesn't support this program type",
 113         [ERRCODE_OFFSET(WRNGPID)]       = "Wrong pid in netlink message",
 114         [ERRCODE_OFFSET(INVSEQ)]        = "Invalid netlink sequence",
 115 };
 116
 117 int libbpf_strerror(int err, char *buf, size_t size)
 118 {
 119         if (!buf || !size)
 120                 return -1;
 121
 122         err = err > 0 ? err : -err;
 123
 124         if (err < __LIBBPF_ERRNO__START) {
 125                 int ret;
 126
 127                 ret = strerror_r(err, buf, size);
 128                 buf[size - 1] = '\0';
 129                 return ret;
 130         }
 131
 132         if (err < __LIBBPF_ERRNO__END) {
 133                 const char *msg;
 134
 135                 msg = libbpf_strerror_table[ERRNO_OFFSET(err)];
 136                 snprintf(buf, size, "%s", msg);
 137                 buf[size - 1] = '\0';
 138                 return 0;
 139         }
 140
 141         snprintf(buf, size, "Unknown libbpf error %d", err);
 142         buf[size - 1] = '\0';
 143         return -1;
 144 }
 145
 146 #define CHECK_ERR(action, err, out) do {        \
 147         err = action;                   \
 148         if (err)                        \
 149                 goto out;               \
 150 } while(0)
 151
 152
 153 /* Copied from tools/perf/util/util.h */
 154 #ifndef zfree
 155 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 156 #endif
 157
 158 #ifndef zclose
 159 # define zclose(fd) ({                  \
 160         int ___err = 0;                 \
 161         if ((fd) >= 0)                  \
 162                 ___err = close((fd));   \
 163         fd = -1;                        \
 164         ___err; })
 165 #endif
 166
 167 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 168 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 169 #else
 170 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 171 #endif
 172
 173 /*
 174  * bpf_prog should be a better name but it has been used in
 175  * linux/filter.h.
 176  */
 177 struct bpf_program {
 178         /* Index in elf obj file, for relocation use. */
 179         int idx;
 180         char *name;
 181         int prog_ifindex;
 182         char *section_name;
 183         struct bpf_insn *insns;
 184         size_t insns_cnt, main_prog_cnt;
 185         enum bpf_prog_type type;
 186
 187         struct reloc_desc {
 188                 enum {
 189                         RELO_LD64,
 190                         RELO_CALL,
 191                 } type;
 192                 int insn_idx;
 193                 union {
 194                         int map_idx;
 195                         int text_off;
 196                 };
 197         } *reloc_desc;
 198         int nr_reloc;
 199
 200         struct {
 201                 int nr;
 202                 int *fds;
 203         } instances;
 204         bpf_program_prep_t preprocessor;
 205
 206         struct bpf_object *obj;
 207         void *priv;
 208         bpf_program_clear_priv_t clear_priv;
 209
 210         enum bpf_attach_type expected_attach_type;
 211 };
 212
 213 struct bpf_map {
 214         int fd;
 215         char *name;
 216         size_t offset;
 217         int map_ifindex;
 218         struct bpf_map_def def;
 219         uint32_t btf_key_type_id;
 220         uint32_t btf_value_type_id;
 221         void *priv;
 222         bpf_map_clear_priv_t clear_priv;
 223 };
 224
 225 static LIST_HEAD(bpf_objects_list);
 226
 227 struct bpf_object {
 228         char license[64];
 229         u32 kern_version;
 230
 231         struct bpf_program *programs;
 232         size_t nr_programs;
 233         struct bpf_map *maps;
 234         size_t nr_maps;
 235
 236         bool loaded;
 237
 238         /*
 239          * Information when doing elf related work. Only valid if fd
 240          * is valid.
 241          */
 242         struct {
 243                 int fd;
 244                 void *obj_buf;
 245                 size_t obj_buf_sz;
 246                 Elf *elf;
 247                 GElf_Ehdr ehdr;
 248                 Elf_Data *symbols;
 249                 size_t strtabidx;
 250                 struct {
 251                         GElf_Shdr shdr;
 252                         Elf_Data *data;
 253                 } *reloc;
 254                 int nr_reloc;
 255                 int maps_shndx;
 256                 int text_shndx;
 257         } efile;
 258         /*
 259          * All loaded bpf_object is linked in a list, which is
 260          * hidden to caller. bpf_objects__<func> handlers deal with
 261          * all objects.
 262          */
 263         struct list_head list;
 264
 265         struct btf *btf;
 266
 267         void *priv;
 268         bpf_object_clear_priv_t clear_priv;
 269
 270         char path[];
 271 };
 272 #define obj_elf_valid(o)        ((o)->efile.elf)
 273
 274 static void bpf_program__unload(struct bpf_program *prog)
 275 {
 276         int i;
 277
 278         if (!prog)
 279                 return;
 280
 281         /*
 282          * If the object is opened but the program was never loaded,
 283          * it is possible that prog->instances.nr == -1.
 284          */
 285         if (prog->instances.nr > 0) {
 286                 for (i = 0; i < prog->instances.nr; i++)
 287                         zclose(prog->instances.fds[i]);
 288         } else if (prog->instances.nr != -1) {
 289                 pr_warning("Internal error: instances.nr is %d\n",
 290                            prog->instances.nr);
 291         }
 292
 293         prog->instances.nr = -1;
 294         zfree(&prog->instances.fds);
 295 }
 296
 297 static void bpf_program__exit(struct bpf_program *prog)
 298 {
 299         if (!prog)
 300                 return;
 301
 302         if (prog->clear_priv)
 303                 prog->clear_priv(prog, prog->priv);
 304
 305         prog->priv = NULL;
 306         prog->clear_priv = NULL;
 307
 308         bpf_program__unload(prog);
 309         zfree(&prog->name);
 310         zfree(&prog->section_name);
 311         zfree(&prog->insns);
 312         zfree(&prog->reloc_desc);
 313
 314         prog->nr_reloc = 0;
 315         prog->insns_cnt = 0;
 316         prog->idx = -1;
 317 }
 318
 319 static int
 320 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 321                   struct bpf_program *prog)
 322 {
 323         if (size < sizeof(struct bpf_insn)) {
 324                 pr_warning("corrupted section '%s'\n", section_name);
 325                 return -EINVAL;
 326         }
 327
 328         bzero(prog, sizeof(*prog));
 329
 330         prog->section_name = strdup(section_name);
 331         if (!prog->section_name) {
 332                 pr_warning("failed to alloc name for prog under section(%d) %s\n",
 333                            idx, section_name);
 334                 goto errout;
 335         }
 336
 337         prog->insns = malloc(size);
 338         if (!prog->insns) {
 339                 pr_warning("failed to alloc insns for prog under section %s\n",
 340                            section_name);
 341                 goto errout;
 342         }
 343         prog->insns_cnt = size / sizeof(struct bpf_insn);
 344         memcpy(prog->insns, data,
 345                prog->insns_cnt * sizeof(struct bpf_insn));
 346         prog->idx = idx;
 347         prog->instances.fds = NULL;
 348         prog->instances.nr = -1;
 349         prog->type = BPF_PROG_TYPE_KPROBE;
 350
 351         return 0;
 352 errout:
 353         bpf_program__exit(prog);
 354         return -ENOMEM;
 355 }
 356
 357 static int
 358 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 359                         char *section_name, int idx)
 360 {
 361         struct bpf_program prog, *progs;
 362         int nr_progs, err;
 363
 364         err = bpf_program__init(data, size, section_name, idx, &prog);
 365         if (err)
 366                 return err;
 367
 368         progs = obj->programs;
 369         nr_progs = obj->nr_programs;
 370
 371         progs = realloc(progs, sizeof(progs[0]) * (nr_progs + 1));
 372         if (!progs) {
 373                 /*
 374                  * In this case the original obj->programs
 375                  * is still valid, so don't need special treat for
 376                  * bpf_close_object().
 377                  */
 378                 pr_warning("failed to alloc a new program under section '%s'\n",
 379                            section_name);
 380                 bpf_program__exit(&prog);
 381                 return -ENOMEM;
 382         }
 383
 384         pr_debug("found program %s\n", prog.section_name);
 385         obj->programs = progs;
 386         obj->nr_programs = nr_progs + 1;
 387         prog.obj = obj;
 388         progs[nr_progs] = prog;
 389         return 0;
 390 }
 391
 392 static int
 393 bpf_object__init_prog_names(struct bpf_object *obj)
 394 {
 395         Elf_Data *symbols = obj->efile.symbols;
 396         struct bpf_program *prog;
 397         size_t pi, si;
 398
 399         for (pi = 0; pi < obj->nr_programs; pi++) {
 400                 const char *name = NULL;
 401
 402                 prog = &obj->programs[pi];
 403                 if (prog->idx == obj->efile.text_shndx) {
 404                         name = ".text";
 405                         goto skip_search;
 406                 }
 407
 408                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 409                      si++) {
 410                         GElf_Sym sym;
 411
 412                         if (!gelf_getsym(symbols, si, &sym))
 413                                 continue;
 414                         if (sym.st_shndx != prog->idx)
 415                                 continue;
 416                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 417                                 continue;
 418
 419                         name = elf_strptr(obj->efile.elf,
 420                                           obj->efile.strtabidx,
 421                                           sym.st_name);
 422                         if (!name) {
 423                                 pr_warning("failed to get sym name string for prog %s\n",
 424                                            prog->section_name);
 425                                 return -LIBBPF_ERRNO__LIBELF;
 426                         }
 427                 }
 428
 429                 if (!name) {
 430                         pr_warning("failed to find sym for prog %s\n",
 431                                    prog->section_name);
 432                         return -EINVAL;
 433                 }
 434 skip_search:
 435                 prog->name = strdup(name);
 436                 if (!prog->name) {
 437                         pr_warning("failed to allocate memory for prog sym %s\n",
 438                                    name);
 439                         return -ENOMEM;
 440                 }
 441         }
 442
 443         return 0;
 444 }
 445
 446 static struct bpf_object *bpf_object__new(const char *path,
 447                                           void *obj_buf,
 448                                           size_t obj_buf_sz)
 449 {
 450         struct bpf_object *obj;
 451
 452         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 453         if (!obj) {
 454                 pr_warning("alloc memory failed for %s\n", path);
 455                 return ERR_PTR(-ENOMEM);
 456         }
 457
 458         strcpy(obj->path, path);
 459         obj->efile.fd = -1;
 460
 461         /*
 462          * Caller of this function should also calls
 463          * bpf_object__elf_finish() after data collection to return
 464          * obj_buf to user. If not, we should duplicate the buffer to
 465          * avoid user freeing them before elf finish.
 466          */
 467         obj->efile.obj_buf = obj_buf;
 468         obj->efile.obj_buf_sz = obj_buf_sz;
 469         obj->efile.maps_shndx = -1;
 470
 471         obj->loaded = false;
 472
 473         INIT_LIST_HEAD(&obj->list);
 474         list_add(&obj->list, &bpf_objects_list);
 475         return obj;
 476 }
 477
 478 static void bpf_object__elf_finish(struct bpf_object *obj)
 479 {
 480         if (!obj_elf_valid(obj))
 481                 return;
 482
 483         if (obj->efile.elf) {
 484                 elf_end(obj->efile.elf);
 485                 obj->efile.elf = NULL;
 486         }
 487         obj->efile.symbols = NULL;
 488
 489         zfree(&obj->efile.reloc);
 490         obj->efile.nr_reloc = 0;
 491         zclose(obj->efile.fd);
 492         obj->efile.obj_buf = NULL;
 493         obj->efile.obj_buf_sz = 0;
 494 }
 495
 496 static int bpf_object__elf_init(struct bpf_object *obj)
 497 {
 498         int err = 0;
 499         GElf_Ehdr *ep;
 500
 501         if (obj_elf_valid(obj)) {
 502                 pr_warning("elf init: internal error\n");
 503                 return -LIBBPF_ERRNO__LIBELF;
 504         }
 505
 506         if (obj->efile.obj_buf_sz > 0) {
 507                 /*
 508                  * obj_buf should have been validated by
 509                  * bpf_object__open_buffer().
 510                  */
 511                 obj->efile.elf = elf_memory(obj->efile.obj_buf,
 512                                             obj->efile.obj_buf_sz);
 513         } else {
 514                 obj->efile.fd = open(obj->path, O_RDONLY);
 515                 if (obj->efile.fd < 0) {
 516                         pr_warning("failed to open %s: %s\n", obj->path,
 517                                         strerror(errno));
 518                         return -errno;
 519                 }
 520
 521                 obj->efile.elf = elf_begin(obj->efile.fd,
 522                                 LIBBPF_ELF_C_READ_MMAP,
 523                                 NULL);
 524         }
 525
 526         if (!obj->efile.elf) {
 527                 pr_warning("failed to open %s as ELF file\n",
 528                                 obj->path);
 529                 err = -LIBBPF_ERRNO__LIBELF;
 530                 goto errout;
 531         }
 532
 533         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 534                 pr_warning("failed to get EHDR from %s\n",
 535                                 obj->path);
 536                 err = -LIBBPF_ERRNO__FORMAT;
 537                 goto errout;
 538         }
 539         ep = &obj->efile.ehdr;
 540
 541         /* Old LLVM set e_machine to EM_NONE */
 542         if ((ep->e_type != ET_REL) || (ep->e_machine && (ep->e_machine != EM_BPF))) {
 543                 pr_warning("%s is not an eBPF object file\n",
 544                         obj->path);
 545                 err = -LIBBPF_ERRNO__FORMAT;
 546                 goto errout;
 547         }
 548
 549         return 0;
 550 errout:
 551         bpf_object__elf_finish(obj);
 552         return err;
 553 }
 554
 555 static int
 556 bpf_object__check_endianness(struct bpf_object *obj)
 557 {
 558         static unsigned int const endian = 1;
 559
 560         switch (obj->efile.ehdr.e_ident[EI_DATA]) {
 561         case ELFDATA2LSB:
 562                 /* We are big endian, BPF obj is little endian. */
 563                 if (*(unsigned char const *)&endian != 1)
 564                         goto mismatch;
 565                 break;
 566
 567         case ELFDATA2MSB:
 568                 /* We are little endian, BPF obj is big endian. */
 569                 if (*(unsigned char const *)&endian != 0)
 570                         goto mismatch;
 571                 break;
 572         default:
 573                 return -LIBBPF_ERRNO__ENDIAN;
 574         }
 575
 576         return 0;
 577
 578 mismatch:
 579         pr_warning("Error: endianness mismatch.\n");
 580         return -LIBBPF_ERRNO__ENDIAN;
 581 }
 582
 583 static int
 584 bpf_object__init_license(struct bpf_object *obj,
 585                          void *data, size_t size)
 586 {
 587         memcpy(obj->license, data,
 588                min(size, sizeof(obj->license) - 1));
 589         pr_debug("license of %s is %s\n", obj->path, obj->license);
 590         return 0;
 591 }
 592
 593 static int
 594 bpf_object__init_kversion(struct bpf_object *obj,
 595                           void *data, size_t size)
 596 {
 597         u32 kver;
 598
 599         if (size != sizeof(kver)) {
 600                 pr_warning("invalid kver section in %s\n", obj->path);
 601                 return -LIBBPF_ERRNO__FORMAT;
 602         }
 603         memcpy(&kver, data, sizeof(kver));
 604         obj->kern_version = kver;
 605         pr_debug("kernel version of %s is %x\n", obj->path,
 606                  obj->kern_version);
 607         return 0;
 608 }
 609
 610 static int compare_bpf_map(const void *_a, const void *_b)
 611 {
 612         const struct bpf_map *a = _a;
 613         const struct bpf_map *b = _b;
 614
 615         return a->offset - b->offset;
 616 }
 617
 618 static int
 619 bpf_object__init_maps(struct bpf_object *obj)
 620 {
 621         int i, map_idx, map_def_sz, nr_maps = 0;
 622         Elf_Scn *scn;
 623         Elf_Data *data;
 624         Elf_Data *symbols = obj->efile.symbols;
 625
 626         if (obj->efile.maps_shndx < 0)
 627                 return -EINVAL;
 628         if (!symbols)
 629                 return -EINVAL;
 630
 631         scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
 632         if (scn)
 633                 data = elf_getdata(scn, NULL);
 634         if (!scn || !data) {
 635                 pr_warning("failed to get Elf_Data from map section %d\n",
 636                            obj->efile.maps_shndx);
 637                 return -EINVAL;
 638         }
 639
 640         /*
 641          * Count number of maps. Each map has a name.
 642          * Array of maps is not supported: only the first element is
 643          * considered.
 644          *
 645          * TODO: Detect array of map and report error.
 646          */
 647         for (i = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 648                 GElf_Sym sym;
 649
 650                 if (!gelf_getsym(symbols, i, &sym))
 651                         continue;
 652                 if (sym.st_shndx != obj->efile.maps_shndx)
 653                         continue;
 654                 nr_maps++;
 655         }
 656
 657         /* Alloc obj->maps and fill nr_maps. */
 658         pr_debug("maps in %s: %d maps in %zd bytes\n", obj->path,
 659                  nr_maps, data->d_size);
 660
 661         if (!nr_maps)
 662                 return 0;
 663
 664         /* Assume equally sized map definitions */
 665         map_def_sz = data->d_size / nr_maps;
 666         if (!data->d_size || (data->d_size % nr_maps) != 0) {
 667                 pr_warning("unable to determine map definition size "
 668                            "section %s, %d maps in %zd bytes\n",
 669                            obj->path, nr_maps, data->d_size);
 670                 return -EINVAL;
 671         }
 672
 673         obj->maps = calloc(nr_maps, sizeof(obj->maps[0]));
 674         if (!obj->maps) {
 675                 pr_warning("alloc maps for object failed\n");
 676                 return -ENOMEM;
 677         }
 678         obj->nr_maps = nr_maps;
 679
 680         /*
 681          * fill all fd with -1 so won't close incorrect
 682          * fd (fd=0 is stdin) when failure (zclose won't close
 683          * negative fd)).
 684          */
 685         for (i = 0; i < nr_maps; i++)
 686                 obj->maps[i].fd = -1;
 687
 688         /*
 689          * Fill obj->maps using data in "maps" section.
 690          */
 691         for (i = 0, map_idx = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 692                 GElf_Sym sym;
 693                 const char *map_name;
 694                 struct bpf_map_def *def;
 695
 696                 if (!gelf_getsym(symbols, i, &sym))
 697                         continue;
 698                 if (sym.st_shndx != obj->efile.maps_shndx)
 699                         continue;
 700
 701                 map_name = elf_strptr(obj->efile.elf,
 702                                       obj->efile.strtabidx,
 703                                       sym.st_name);
 704                 obj->maps[map_idx].offset = sym.st_value;
 705                 if (sym.st_value + map_def_sz > data->d_size) {
 706                         pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
 707                                    obj->path, map_name);
 708                         return -EINVAL;
 709                 }
 710
 711                 obj->maps[map_idx].name = strdup(map_name);
 712                 if (!obj->maps[map_idx].name) {
 713                         pr_warning("failed to alloc map name\n");
 714                         return -ENOMEM;
 715                 }
 716                 pr_debug("map %d is \"%s\"\n", map_idx,
 717                          obj->maps[map_idx].name);
 718                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
 719                 /*
 720                  * If the definition of the map in the object file fits in
 721                  * bpf_map_def, copy it.  Any extra fields in our version
 722                  * of bpf_map_def will default to zero as a result of the
 723                  * calloc above.
 724                  */
 725                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
 726                         memcpy(&obj->maps[map_idx].def, def, map_def_sz);
 727                 } else {
 728                         /*
 729                          * Here the map structure being read is bigger than what
 730                          * we expect, truncate if the excess bits are all zero.
 731                          * If they are not zero, reject this map as
 732                          * incompatible.
 733                          */
 734                         char *b;
 735                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
 736                              b < ((char *)def) + map_def_sz; b++) {
 737                                 if (*b != 0) {
 738                                         pr_warning("maps section in %s: \"%s\" "
 739                                                    "has unrecognized, non-zero "
 740                                                    "options\n",
 741                                                    obj->path, map_name);
 742                                         return -EINVAL;
 743                                 }
 744                         }
 745                         memcpy(&obj->maps[map_idx].def, def,
 746                                sizeof(struct bpf_map_def));
 747                 }
 748                 map_idx++;
 749         }
 750
 751         qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]), compare_bpf_map);
 752         return 0;
 753 }
 754
 755 static bool section_have_execinstr(struct bpf_object *obj, int idx)
 756 {
 757         Elf_Scn *scn;
 758         GElf_Shdr sh;
 759
 760         scn = elf_getscn(obj->efile.elf, idx);
 761         if (!scn)
 762                 return false;
 763
 764         if (gelf_getshdr(scn, &sh) != &sh)
 765                 return false;
 766
 767         if (sh.sh_flags & SHF_EXECINSTR)
 768                 return true;
 769
 770         return false;
 771 }
 772
 773 static int bpf_object__elf_collect(struct bpf_object *obj)
 774 {
 775         Elf *elf = obj->efile.elf;
 776         GElf_Ehdr *ep = &obj->efile.ehdr;
 777         Elf_Scn *scn = NULL;
 778         int idx = 0, err = 0;
 779
 780         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
 781         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
 782                 pr_warning("failed to get e_shstrndx from %s\n",
 783                            obj->path);
 784                 return -LIBBPF_ERRNO__FORMAT;
 785         }
 786
 787         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 788                 char *name;
 789                 GElf_Shdr sh;
 790                 Elf_Data *data;
 791
 792                 idx++;
 793                 if (gelf_getshdr(scn, &sh) != &sh) {
 794                         pr_warning("failed to get section(%d) header from %s\n",
 795                                    idx, obj->path);
 796                         err = -LIBBPF_ERRNO__FORMAT;
 797                         goto out;
 798                 }
 799
 800                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 801                 if (!name) {
 802                         pr_warning("failed to get section(%d) name from %s\n",
 803                                    idx, obj->path);
 804                         err = -LIBBPF_ERRNO__FORMAT;
 805                         goto out;
 806                 }
 807
 808                 data = elf_getdata(scn, 0);
 809                 if (!data) {
 810                         pr_warning("failed to get section(%d) data from %s(%s)\n",
 811                                    idx, name, obj->path);
 812                         err = -LIBBPF_ERRNO__FORMAT;
 813                         goto out;
 814                 }
 815                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
 816                          idx, name, (unsigned long)data->d_size,
 817                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
 818                          (int)sh.sh_type);
 819
 820                 if (strcmp(name, "license") == 0)
 821                         err = bpf_object__init_license(obj,
 822                                                        data->d_buf,
 823                                                        data->d_size);
 824                 else if (strcmp(name, "version") == 0)
 825                         err = bpf_object__init_kversion(obj,
 826                                                         data->d_buf,
 827                                                         data->d_size);
 828                 else if (strcmp(name, "maps") == 0)
 829                         obj->efile.maps_shndx = idx;
 830                 else if (strcmp(name, BTF_ELF_SEC) == 0) {
 831                         obj->btf = btf__new(data->d_buf, data->d_size,
 832                                             __pr_debug);
 833                         if (IS_ERR(obj->btf)) {
 834                                 pr_warning("Error loading ELF section %s: %ld. Ignored and continue.\n",
 835                                            BTF_ELF_SEC, PTR_ERR(obj->btf));
 836                                 obj->btf = NULL;
 837                         }
 838                 } else if (sh.sh_type == SHT_SYMTAB) {
 839                         if (obj->efile.symbols) {
 840                                 pr_warning("bpf: multiple SYMTAB in %s\n",
 841                                            obj->path);
 842                                 err = -LIBBPF_ERRNO__FORMAT;
 843                         } else {
 844                                 obj->efile.symbols = data;
 845                                 obj->efile.strtabidx = sh.sh_link;
 846                         }
 847                 } else if ((sh.sh_type == SHT_PROGBITS) &&
 848                            (sh.sh_flags & SHF_EXECINSTR) &&
 849                            (data->d_size > 0)) {
 850                         if (strcmp(name, ".text") == 0)
 851                                 obj->efile.text_shndx = idx;
 852                         err = bpf_object__add_program(obj, data->d_buf,
 853                                                       data->d_size, name, idx);
 854                         if (err) {
 855                                 char errmsg[STRERR_BUFSIZE];
 856
 857                                 strerror_r(-err, errmsg, sizeof(errmsg));
 858                                 pr_warning("failed to alloc program %s (%s): %s",
 859                                            name, obj->path, errmsg);
 860                         }
 861                 } else if (sh.sh_type == SHT_REL) {
 862                         void *reloc = obj->efile.reloc;
 863                         int nr_reloc = obj->efile.nr_reloc + 1;
 864                         int sec = sh.sh_info; /* points to other section */
 865
 866                         /* Only do relo for section with exec instructions */
 867                         if (!section_have_execinstr(obj, sec)) {
 868                                 pr_debug("skip relo %s(%d) for section(%d)\n",
 869                                          name, idx, sec);
 870                                 continue;
 871                         }
 872
 873                         reloc = realloc(reloc,
 874                                         sizeof(*obj->efile.reloc) * nr_reloc);
 875                         if (!reloc) {
 876                                 pr_warning("realloc failed\n");
 877                                 err = -ENOMEM;
 878                         } else {
 879                                 int n = nr_reloc - 1;
 880
 881                                 obj->efile.reloc = reloc;
 882                                 obj->efile.nr_reloc = nr_reloc;
 883
 884                                 obj->efile.reloc[n].shdr = sh;
 885                                 obj->efile.reloc[n].data = data;
 886                         }
 887                 } else {
 888                         pr_debug("skip section(%d) %s\n", idx, name);
 889                 }
 890                 if (err)
 891                         goto out;
 892         }
 893
 894         if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
 895                 pr_warning("Corrupted ELF file: index of strtab invalid\n");
 896                 return LIBBPF_ERRNO__FORMAT;
 897         }
 898         if (obj->efile.maps_shndx >= 0) {
 899                 err = bpf_object__init_maps(obj);
 900                 if (err)
 901                         goto out;
 902         }
 903         err = bpf_object__init_prog_names(obj);
 904 out:
 905         return err;
 906 }
 907
 908 static struct bpf_program *
 909 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
 910 {
 911         struct bpf_program *prog;
 912         size_t i;
 913
 914         for (i = 0; i < obj->nr_programs; i++) {
 915                 prog = &obj->programs[i];
 916                 if (prog->idx == idx)
 917                         return prog;
 918         }
 919         return NULL;
 920 }
 921
 922 static int
 923 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
 924                            Elf_Data *data, struct bpf_object *obj)
 925 {
 926         Elf_Data *symbols = obj->efile.symbols;
 927         int text_shndx = obj->efile.text_shndx;
 928         int maps_shndx = obj->efile.maps_shndx;
 929         struct bpf_map *maps = obj->maps;
 930         size_t nr_maps = obj->nr_maps;
 931         int i, nrels;
 932
 933         pr_debug("collecting relocating info for: '%s'\n",
 934                  prog->section_name);
 935         nrels = shdr->sh_size / shdr->sh_entsize;
 936
 937         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
 938         if (!prog->reloc_desc) {
 939                 pr_warning("failed to alloc memory in relocation\n");
 940                 return -ENOMEM;
 941         }
 942         prog->nr_reloc = nrels;
 943
 944         for (i = 0; i < nrels; i++) {
 945                 GElf_Sym sym;
 946                 GElf_Rel rel;
 947                 unsigned int insn_idx;
 948                 struct bpf_insn *insns = prog->insns;
 949                 size_t map_idx;
 950
 951                 if (!gelf_getrel(data, i, &rel)) {
 952                         pr_warning("relocation: failed to get %d reloc\n", i);
 953                         return -LIBBPF_ERRNO__FORMAT;
 954                 }
 955
 956                 if (!gelf_getsym(symbols,
 957                                  GELF_R_SYM(rel.r_info),
 958                                  &sym)) {
 959                         pr_warning("relocation: symbol %"PRIx64" not found\n",
 960                                    GELF_R_SYM(rel.r_info));
 961                         return -LIBBPF_ERRNO__FORMAT;
 962                 }
 963                 pr_debug("relo for %lld value %lld name %d\n",
 964                          (long long) (rel.r_info >> 32),
 965                          (long long) sym.st_value, sym.st_name);
 966
 967                 if (sym.st_shndx != maps_shndx && sym.st_shndx != text_shndx) {
 968                         pr_warning("Program '%s' contains non-map related relo data pointing to section %u\n",
 969                                    prog->section_name, sym.st_shndx);
 970                         return -LIBBPF_ERRNO__RELOC;
 971                 }
 972
 973                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
 974                 pr_debug("relocation: insn_idx=%u\n", insn_idx);
 975
 976                 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
 977                         if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
 978                                 pr_warning("incorrect bpf_call opcode\n");
 979                                 return -LIBBPF_ERRNO__RELOC;
 980                         }
 981                         prog->reloc_desc[i].type = RELO_CALL;
 982                         prog->reloc_desc[i].insn_idx = insn_idx;
 983                         prog->reloc_desc[i].text_off = sym.st_value;
 984                         continue;
 985                 }
 986
 987                 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
 988                         pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
 989                                    insn_idx, insns[insn_idx].code);
 990                         return -LIBBPF_ERRNO__RELOC;
 991                 }
 992
 993                 /* TODO: 'maps' is sorted. We can use bsearch to make it faster. */
 994                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
 995                         if (maps[map_idx].offset == sym.st_value) {
 996                                 pr_debug("relocation: find map %zd (%s) for insn %u\n",
 997                                          map_idx, maps[map_idx].name, insn_idx);
 998                                 break;
 999                         }
1000                 }
1001
1002                 if (map_idx >= nr_maps) {
1003                         pr_warning("bpf relocation: map_idx %d large than %d\n",
1004                                    (int)map_idx, (int)nr_maps - 1);
1005                         return -LIBBPF_ERRNO__RELOC;
1006                 }
1007
1008                 prog->reloc_desc[i].type = RELO_LD64;
1009                 prog->reloc_desc[i].insn_idx = insn_idx;
1010                 prog->reloc_desc[i].map_idx = map_idx;
1011         }
1012         return 0;
1013 }
1014
1015 static int bpf_map_find_btf_info(struct bpf_map *map, const struct btf *btf)
1016 {
1017         struct bpf_map_def *def = &map->def;
1018         const size_t max_name = 256;
1019         int64_t key_size, value_size;
1020         int32_t key_id, value_id;
1021         char name[max_name];
1022
1023         /* Find key type by name from BTF */
1024         if (snprintf(name, max_name, "%s_key", map->name) == max_name) {
1025                 pr_warning("map:%s length of BTF key_type:%s_key is too long\n",
1026                            map->name, map->name);
1027                 return -EINVAL;
1028         }
1029
1030         key_id = btf__find_by_name(btf, name);
1031         if (key_id < 0) {
1032                 pr_debug("map:%s key_type:%s cannot be found in BTF\n",
1033                          map->name, name);
1034                 return key_id;
1035         }
1036
1037         key_size = btf__resolve_size(btf, key_id);
1038         if (key_size < 0) {
1039                 pr_warning("map:%s key_type:%s cannot get the BTF type_size\n",
1040                            map->name, name);
1041                 return key_size;
1042         }
1043
1044         if (def->key_size != key_size) {
1045                 pr_warning("map:%s key_type:%s has BTF type_size:%u != key_size:%u\n",
1046                            map->name, name, (unsigned int)key_size, def->key_size);
1047                 return -EINVAL;
1048         }
1049
1050         /* Find value type from BTF */
1051         if (snprintf(name, max_name, "%s_value", map->name) == max_name) {
1052                 pr_warning("map:%s length of BTF value_type:%s_value is too long\n",
1053                           map->name, map->name);
1054                 return -EINVAL;
1055         }
1056
1057         value_id = btf__find_by_name(btf, name);
1058         if (value_id < 0) {
1059                 pr_debug("map:%s value_type:%s cannot be found in BTF\n",
1060                          map->name, name);
1061                 return value_id;
1062         }
1063
1064         value_size = btf__resolve_size(btf, value_id);
1065         if (value_size < 0) {
1066                 pr_warning("map:%s value_type:%s cannot get the BTF type_size\n",
1067                            map->name, name);
1068                 return value_size;
1069         }
1070
1071         if (def->value_size != value_size) {
1072                 pr_warning("map:%s value_type:%s has BTF type_size:%u != value_size:%u\n",
1073                            map->name, name, (unsigned int)value_size, def->value_size);
1074                 return -EINVAL;
1075         }
1076
1077         map->btf_key_type_id = key_id;
1078         map->btf_value_type_id = value_id;
1079
1080         return 0;
1081 }
1082
1083 static int
1084 bpf_object__create_maps(struct bpf_object *obj)
1085 {
1086         struct bpf_create_map_attr create_attr = {};
1087         unsigned int i;
1088         int err;
1089
1090         for (i = 0; i < obj->nr_maps; i++) {
1091                 struct bpf_map *map = &obj->maps[i];
1092                 struct bpf_map_def *def = &map->def;
1093                 int *pfd = &map->fd;
1094
1095                 create_attr.name = map->name;
1096                 create_attr.map_ifindex = map->map_ifindex;
1097                 create_attr.map_type = def->type;
1098                 create_attr.map_flags = def->map_flags;
1099                 create_attr.key_size = def->key_size;
1100                 create_attr.value_size = def->value_size;
1101                 create_attr.max_entries = def->max_entries;
1102                 create_attr.btf_fd = 0;
1103                 create_attr.btf_key_type_id = 0;
1104                 create_attr.btf_value_type_id = 0;
1105
1106                 if (obj->btf && !bpf_map_find_btf_info(map, obj->btf)) {
1107                         create_attr.btf_fd = btf__fd(obj->btf);
1108                         create_attr.btf_key_type_id = map->btf_key_type_id;
1109                         create_attr.btf_value_type_id = map->btf_value_type_id;
1110                 }
1111
1112                 *pfd = bpf_create_map_xattr(&create_attr);
1113                 if (*pfd < 0 && create_attr.btf_key_type_id) {
1114                         pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
1115                                    map->name, strerror(errno), errno);
1116                         create_attr.btf_fd = 0;
1117                         create_attr.btf_key_type_id = 0;
1118                         create_attr.btf_value_type_id = 0;
1119                         map->btf_key_type_id = 0;
1120                         map->btf_value_type_id = 0;
1121                         *pfd = bpf_create_map_xattr(&create_attr);
1122                 }
1123
1124                 if (*pfd < 0) {
1125                         size_t j;
1126
1127                         err = *pfd;
1128                         pr_warning("failed to create map (name: '%s'): %s\n",
1129                                    map->name,
1130                                    strerror(errno));
1131                         for (j = 0; j < i; j++)
1132                                 zclose(obj->maps[j].fd);
1133                         return err;
1134                 }
1135                 pr_debug("create map %s: fd=%d\n", map->name, *pfd);
1136         }
1137
1138         return 0;
1139 }
1140
1141 static int
1142 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
1143                         struct reloc_desc *relo)
1144 {
1145         struct bpf_insn *insn, *new_insn;
1146         struct bpf_program *text;
1147         size_t new_cnt;
1148
1149         if (relo->type != RELO_CALL)
1150                 return -LIBBPF_ERRNO__RELOC;
1151
1152         if (prog->idx == obj->efile.text_shndx) {
1153                 pr_warning("relo in .text insn %d into off %d\n",
1154                            relo->insn_idx, relo->text_off);
1155                 return -LIBBPF_ERRNO__RELOC;
1156         }
1157
1158         if (prog->main_prog_cnt == 0) {
1159                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
1160                 if (!text) {
1161                         pr_warning("no .text section found yet relo into text exist\n");
1162                         return -LIBBPF_ERRNO__RELOC;
1163                 }
1164                 new_cnt = prog->insns_cnt + text->insns_cnt;
1165                 new_insn = realloc(prog->insns, new_cnt * sizeof(*insn));
1166                 if (!new_insn) {
1167                         pr_warning("oom in prog realloc\n");
1168                         return -ENOMEM;
1169                 }
1170                 memcpy(new_insn + prog->insns_cnt, text->insns,
1171                        text->insns_cnt * sizeof(*insn));
1172                 prog->insns = new_insn;
1173                 prog->main_prog_cnt = prog->insns_cnt;
1174                 prog->insns_cnt = new_cnt;
1175                 pr_debug("added %zd insn from %s to prog %s\n",
1176                          text->insns_cnt, text->section_name,
1177                          prog->section_name);
1178         }
1179         insn = &prog->insns[relo->insn_idx];
1180         insn->imm += prog->main_prog_cnt - relo->insn_idx;
1181         return 0;
1182 }
1183
1184 static int
1185 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
1186 {
1187         int i, err;
1188
1189         if (!prog || !prog->reloc_desc)
1190                 return 0;
1191
1192         for (i = 0; i < prog->nr_reloc; i++) {
1193                 if (prog->reloc_desc[i].type == RELO_LD64) {
1194                         struct bpf_insn *insns = prog->insns;
1195                         int insn_idx, map_idx;
1196
1197                         insn_idx = prog->reloc_desc[i].insn_idx;
1198                         map_idx = prog->reloc_desc[i].map_idx;
1199
1200                         if (insn_idx >= (int)prog->insns_cnt) {
1201                                 pr_warning("relocation out of range: '%s'\n",
1202                                            prog->section_name);
1203                                 return -LIBBPF_ERRNO__RELOC;
1204                         }
1205                         insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
1206                         insns[insn_idx].imm = obj->maps[map_idx].fd;
1207                 } else {
1208                         err = bpf_program__reloc_text(prog, obj,
1209                                                       &prog->reloc_desc[i]);
1210                         if (err)
1211                                 return err;
1212                 }
1213         }
1214
1215         zfree(&prog->reloc_desc);
1216         prog->nr_reloc = 0;
1217         return 0;
1218 }
1219
1220
1221 static int
1222 bpf_object__relocate(struct bpf_object *obj)
1223 {
1224         struct bpf_program *prog;
1225         size_t i;
1226         int err;
1227
1228         for (i = 0; i < obj->nr_programs; i++) {
1229                 prog = &obj->programs[i];
1230
1231                 err = bpf_program__relocate(prog, obj);
1232                 if (err) {
1233                         pr_warning("failed to relocate '%s'\n",
1234                                    prog->section_name);
1235                         return err;
1236                 }
1237         }
1238         return 0;
1239 }
1240
1241 static int bpf_object__collect_reloc(struct bpf_object *obj)
1242 {
1243         int i, err;
1244
1245         if (!obj_elf_valid(obj)) {
1246                 pr_warning("Internal error: elf object is closed\n");
1247                 return -LIBBPF_ERRNO__INTERNAL;
1248         }
1249
1250         for (i = 0; i < obj->efile.nr_reloc; i++) {
1251                 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
1252                 Elf_Data *data = obj->efile.reloc[i].data;
1253                 int idx = shdr->sh_info;
1254                 struct bpf_program *prog;
1255
1256                 if (shdr->sh_type != SHT_REL) {
1257                         pr_warning("internal error at %d\n", __LINE__);
1258                         return -LIBBPF_ERRNO__INTERNAL;
1259                 }
1260
1261                 prog = bpf_object__find_prog_by_idx(obj, idx);
1262                 if (!prog) {
1263                         pr_warning("relocation failed: no section(%d)\n", idx);
1264                         return -LIBBPF_ERRNO__RELOC;
1265                 }
1266
1267                 err = bpf_program__collect_reloc(prog,
1268                                                  shdr, data,
1269                                                  obj);
1270                 if (err)
1271                         return err;
1272         }
1273         return 0;
1274 }
1275
1276 static int
1277 load_program(enum bpf_prog_type type, enum bpf_attach_type expected_attach_type,
1278              const char *name, struct bpf_insn *insns, int insns_cnt,
1279              char *license, u32 kern_version, int *pfd, int prog_ifindex)
1280 {
1281         struct bpf_load_program_attr load_attr;
1282         char *log_buf;
1283         int ret;
1284
1285         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
1286         load_attr.prog_type = type;
1287         load_attr.expected_attach_type = expected_attach_type;
1288         load_attr.name = name;
1289         load_attr.insns = insns;
1290         load_attr.insns_cnt = insns_cnt;
1291         load_attr.license = license;
1292         load_attr.kern_version = kern_version;
1293         load_attr.prog_ifindex = prog_ifindex;
1294
1295         if (!load_attr.insns || !load_attr.insns_cnt)
1296                 return -EINVAL;
1297
1298         log_buf = malloc(BPF_LOG_BUF_SIZE);
1299         if (!log_buf)
1300                 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
1301
1302         ret = bpf_load_program_xattr(&load_attr, log_buf, BPF_LOG_BUF_SIZE);
1303
1304         if (ret >= 0) {
1305                 *pfd = ret;
1306                 ret = 0;
1307                 goto out;
1308         }
1309
1310         ret = -LIBBPF_ERRNO__LOAD;
1311         pr_warning("load bpf program failed: %s\n", strerror(errno));
1312
1313         if (log_buf && log_buf[0] != '\0') {
1314                 ret = -LIBBPF_ERRNO__VERIFY;
1315                 pr_warning("-- BEGIN DUMP LOG ---\n");
1316                 pr_warning("\n%s\n", log_buf);
1317                 pr_warning("-- END LOG --\n");
1318         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
1319                 pr_warning("Program too large (%zu insns), at most %d insns\n",
1320                            load_attr.insns_cnt, BPF_MAXINSNS);
1321                 ret = -LIBBPF_ERRNO__PROG2BIG;
1322         } else {
1323                 /* Wrong program type? */
1324                 if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
1325                         int fd;
1326
1327                         load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
1328                         load_attr.expected_attach_type = 0;
1329                         fd = bpf_load_program_xattr(&load_attr, NULL, 0);
1330                         if (fd >= 0) {
1331                                 close(fd);
1332                                 ret = -LIBBPF_ERRNO__PROGTYPE;
1333                                 goto out;
1334                         }
1335                 }
1336
1337                 if (log_buf)
1338                         ret = -LIBBPF_ERRNO__KVER;
1339         }
1340
1341 out:
1342         free(log_buf);
1343         return ret;
1344 }
1345
1346 static int
1347 bpf_program__load(struct bpf_program *prog,
1348                   char *license, u32 kern_version)
1349 {
1350         int err = 0, fd, i;
1351
1352         if (prog->instances.nr < 0 || !prog->instances.fds) {
1353                 if (prog->preprocessor) {
1354                         pr_warning("Internal error: can't load program '%s'\n",
1355                                    prog->section_name);
1356                         return -LIBBPF_ERRNO__INTERNAL;
1357                 }
1358
1359                 prog->instances.fds = malloc(sizeof(int));
1360                 if (!prog->instances.fds) {
1361                         pr_warning("Not enough memory for BPF fds\n");
1362                         return -ENOMEM;
1363                 }
1364                 prog->instances.nr = 1;
1365                 prog->instances.fds[0] = -1;
1366         }
1367
1368         if (!prog->preprocessor) {
1369                 if (prog->instances.nr != 1) {
1370                         pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
1371                                    prog->section_name, prog->instances.nr);
1372                 }
1373                 err = load_program(prog->type, prog->expected_attach_type,
1374                                    prog->name, prog->insns, prog->insns_cnt,
1375                                    license, kern_version, &fd,
1376                                    prog->prog_ifindex);
1377                 if (!err)
1378                         prog->instances.fds[0] = fd;
1379                 goto out;
1380         }
1381
1382         for (i = 0; i < prog->instances.nr; i++) {
1383                 struct bpf_prog_prep_result result;
1384                 bpf_program_prep_t preprocessor = prog->preprocessor;
1385
1386                 bzero(&result, sizeof(result));
1387                 err = preprocessor(prog, i, prog->insns,
1388                                    prog->insns_cnt, &result);
1389                 if (err) {
1390                         pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
1391                                    i, prog->section_name);
1392                         goto out;
1393                 }
1394
1395                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
1396                         pr_debug("Skip loading the %dth instance of program '%s'\n",
1397                                  i, prog->section_name);
1398                         prog->instances.fds[i] = -1;
1399                         if (result.pfd)
1400                                 *result.pfd = -1;
1401                         continue;
1402                 }
1403
1404                 err = load_program(prog->type, prog->expected_attach_type,
1405                                    prog->name, result.new_insn_ptr,
1406                                    result.new_insn_cnt,
1407                                    license, kern_version, &fd,
1408                                    prog->prog_ifindex);
1409
1410                 if (err) {
1411                         pr_warning("Loading the %dth instance of program '%s' failed\n",
1412                                         i, prog->section_name);
1413                         goto out;
1414                 }
1415
1416                 if (result.pfd)
1417                         *result.pfd = fd;
1418                 prog->instances.fds[i] = fd;
1419         }
1420 out:
1421         if (err)
1422                 pr_warning("failed to load program '%s'\n",
1423                            prog->section_name);
1424         zfree(&prog->insns);
1425         prog->insns_cnt = 0;
1426         return err;
1427 }
1428
1429 static int
1430 bpf_object__load_progs(struct bpf_object *obj)
1431 {
1432         size_t i;
1433         int err;
1434
1435         for (i = 0; i < obj->nr_programs; i++) {
1436                 if (obj->programs[i].idx == obj->efile.text_shndx)
1437                         continue;
1438                 err = bpf_program__load(&obj->programs[i],
1439                                         obj->license,
1440                                         obj->kern_version);
1441                 if (err)
1442                         return err;
1443         }
1444         return 0;
1445 }
1446
1447 static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
1448 {
1449         switch (type) {
1450         case BPF_PROG_TYPE_SOCKET_FILTER:
1451         case BPF_PROG_TYPE_SCHED_CLS:
1452         case BPF_PROG_TYPE_SCHED_ACT:
1453         case BPF_PROG_TYPE_XDP:
1454         case BPF_PROG_TYPE_CGROUP_SKB:
1455         case BPF_PROG_TYPE_CGROUP_SOCK:
1456         case BPF_PROG_TYPE_LWT_IN:
1457         case BPF_PROG_TYPE_LWT_OUT:
1458         case BPF_PROG_TYPE_LWT_XMIT:
1459         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
1460         case BPF_PROG_TYPE_SOCK_OPS:
1461         case BPF_PROG_TYPE_SK_SKB:
1462         case BPF_PROG_TYPE_CGROUP_DEVICE:
1463         case BPF_PROG_TYPE_SK_MSG:
1464         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1465         case BPF_PROG_TYPE_LIRC_MODE2:
1466                 return false;
1467         case BPF_PROG_TYPE_UNSPEC:
1468         case BPF_PROG_TYPE_KPROBE:
1469         case BPF_PROG_TYPE_TRACEPOINT:
1470         case BPF_PROG_TYPE_PERF_EVENT:
1471         case BPF_PROG_TYPE_RAW_TRACEPOINT:
1472         default:
1473                 return true;
1474         }
1475 }
1476
1477 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
1478 {
1479         if (needs_kver && obj->kern_version == 0) {
1480                 pr_warning("%s doesn't provide kernel version\n",
1481                            obj->path);
1482                 return -LIBBPF_ERRNO__KVERSION;
1483         }
1484         return 0;
1485 }
1486
1487 static struct bpf_object *
1488 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
1489                    bool needs_kver)
1490 {
1491         struct bpf_object *obj;
1492         int err;
1493
1494         if (elf_version(EV_CURRENT) == EV_NONE) {
1495                 pr_warning("failed to init libelf for %s\n", path);
1496                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1497         }
1498
1499         obj = bpf_object__new(path, obj_buf, obj_buf_sz);
1500         if (IS_ERR(obj))
1501                 return obj;
1502
1503         CHECK_ERR(bpf_object__elf_init(obj), err, out);
1504         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
1505         CHECK_ERR(bpf_object__elf_collect(obj), err, out);
1506         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
1507         CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
1508
1509         bpf_object__elf_finish(obj);
1510         return obj;
1511 out:
1512         bpf_object__close(obj);
1513         return ERR_PTR(err);
1514 }
1515
1516 struct bpf_object *bpf_object__open(const char *path)
1517 {
1518         /* param validation */
1519         if (!path)
1520                 return NULL;
1521
1522         pr_debug("loading %s\n", path);
1523
1524         return __bpf_object__open(path, NULL, 0, true);
1525 }
1526
1527 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
1528                                            size_t obj_buf_sz,
1529                                            const char *name)
1530 {
1531         char tmp_name[64];
1532
1533         /* param validation */
1534         if (!obj_buf || obj_buf_sz <= 0)
1535                 return NULL;
1536
1537         if (!name) {
1538                 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
1539                          (unsigned long)obj_buf,
1540                          (unsigned long)obj_buf_sz);
1541                 tmp_name[sizeof(tmp_name) - 1] = '\0';
1542                 name = tmp_name;
1543         }
1544         pr_debug("loading object '%s' from buffer\n",
1545                  name);
1546
1547         return __bpf_object__open(name, obj_buf, obj_buf_sz, true);
1548 }
1549
1550 int bpf_object__unload(struct bpf_object *obj)
1551 {
1552         size_t i;
1553
1554         if (!obj)
1555                 return -EINVAL;
1556
1557         for (i = 0; i < obj->nr_maps; i++)
1558                 zclose(obj->maps[i].fd);
1559
1560         for (i = 0; i < obj->nr_programs; i++)
1561                 bpf_program__unload(&obj->programs[i]);
1562
1563         return 0;
1564 }
1565
1566 int bpf_object__load(struct bpf_object *obj)
1567 {
1568         int err;
1569
1570         if (!obj)
1571                 return -EINVAL;
1572
1573         if (obj->loaded) {
1574                 pr_warning("object should not be loaded twice\n");
1575                 return -EINVAL;
1576         }
1577
1578         obj->loaded = true;
1579
1580         CHECK_ERR(bpf_object__create_maps(obj), err, out);
1581         CHECK_ERR(bpf_object__relocate(obj), err, out);
1582         CHECK_ERR(bpf_object__load_progs(obj), err, out);
1583
1584         return 0;
1585 out:
1586         bpf_object__unload(obj);
1587         pr_warning("failed to load object '%s'\n", obj->path);
1588         return err;
1589 }
1590
1591 static int check_path(const char *path)
1592 {
1593         struct statfs st_fs;
1594         char *dname, *dir;
1595         int err = 0;
1596
1597         if (path == NULL)
1598                 return -EINVAL;
1599
1600         dname = strdup(path);
1601         if (dname == NULL)
1602                 return -ENOMEM;
1603
1604         dir = dirname(dname);
1605         if (statfs(dir, &st_fs)) {
1606                 pr_warning("failed to statfs %s: %s\n", dir, strerror(errno));
1607                 err = -errno;
1608         }
1609         free(dname);
1610
1611         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
1612                 pr_warning("specified path %s is not on BPF FS\n", path);
1613                 err = -EINVAL;
1614         }
1615
1616         return err;
1617 }
1618
1619 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
1620                               int instance)
1621 {
1622         int err;
1623
1624         err = check_path(path);
1625         if (err)
1626                 return err;
1627
1628         if (prog == NULL) {
1629                 pr_warning("invalid program pointer\n");
1630                 return -EINVAL;
1631         }
1632
1633         if (instance < 0 || instance >= prog->instances.nr) {
1634                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
1635                            instance, prog->section_name, prog->instances.nr);
1636                 return -EINVAL;
1637         }
1638
1639         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
1640                 pr_warning("failed to pin program: %s\n", strerror(errno));
1641                 return -errno;
1642         }
1643         pr_debug("pinned program '%s'\n", path);
1644
1645         return 0;
1646 }
1647
1648 static int make_dir(const char *path)
1649 {
1650         int err = 0;
1651
1652         if (mkdir(path, 0700) && errno != EEXIST)
1653                 err = -errno;
1654
1655         if (err)
1656                 pr_warning("failed to mkdir %s: %s\n", path, strerror(-err));
1657         return err;
1658 }
1659
1660 int bpf_program__pin(struct bpf_program *prog, const char *path)
1661 {
1662         int i, err;
1663
1664         err = check_path(path);
1665         if (err)
1666                 return err;
1667
1668         if (prog == NULL) {
1669                 pr_warning("invalid program pointer\n");
1670                 return -EINVAL;
1671         }
1672
1673         if (prog->instances.nr <= 0) {
1674                 pr_warning("no instances of prog %s to pin\n",
1675                            prog->section_name);
1676                 return -EINVAL;
1677         }
1678
1679         err = make_dir(path);
1680         if (err)
1681                 return err;
1682
1683         for (i = 0; i < prog->instances.nr; i++) {
1684                 char buf[PATH_MAX];
1685                 int len;
1686
1687                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
1688                 if (len < 0)
1689                         return -EINVAL;
1690                 else if (len >= PATH_MAX)
1691                         return -ENAMETOOLONG;
1692
1693                 err = bpf_program__pin_instance(prog, buf, i);
1694                 if (err)
1695                         return err;
1696         }
1697
1698         return 0;
1699 }
1700
1701 int bpf_map__pin(struct bpf_map *map, const char *path)
1702 {
1703         int err;
1704
1705         err = check_path(path);
1706         if (err)
1707                 return err;
1708
1709         if (map == NULL) {
1710                 pr_warning("invalid map pointer\n");
1711                 return -EINVAL;
1712         }
1713
1714         if (bpf_obj_pin(map->fd, path)) {
1715                 pr_warning("failed to pin map: %s\n", strerror(errno));
1716                 return -errno;
1717         }
1718
1719         pr_debug("pinned map '%s'\n", path);
1720         return 0;
1721 }
1722
1723 int bpf_object__pin(struct bpf_object *obj, const char *path)
1724 {
1725         struct bpf_program *prog;
1726         struct bpf_map *map;
1727         int err;
1728
1729         if (!obj)
1730                 return -ENOENT;
1731
1732         if (!obj->loaded) {
1733                 pr_warning("object not yet loaded; load it first\n");
1734                 return -ENOENT;
1735         }
1736
1737         err = make_dir(path);
1738         if (err)
1739                 return err;
1740
1741         bpf_map__for_each(map, obj) {
1742                 char buf[PATH_MAX];
1743                 int len;
1744
1745                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1746                                bpf_map__name(map));
1747                 if (len < 0)
1748                         return -EINVAL;
1749                 else if (len >= PATH_MAX)
1750                         return -ENAMETOOLONG;
1751
1752                 err = bpf_map__pin(map, buf);
1753                 if (err)
1754                         return err;
1755         }
1756
1757         bpf_object__for_each_program(prog, obj) {
1758                 char buf[PATH_MAX];
1759                 int len;
1760
1761                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1762                                prog->section_name);
1763                 if (len < 0)
1764                         return -EINVAL;
1765                 else if (len >= PATH_MAX)
1766                         return -ENAMETOOLONG;
1767
1768                 err = bpf_program__pin(prog, buf);
1769                 if (err)
1770                         return err;
1771         }
1772
1773         return 0;
1774 }
1775
1776 void bpf_object__close(struct bpf_object *obj)
1777 {
1778         size_t i;
1779
1780         if (!obj)
1781                 return;
1782
1783         if (obj->clear_priv)
1784                 obj->clear_priv(obj, obj->priv);
1785
1786         bpf_object__elf_finish(obj);
1787         bpf_object__unload(obj);
1788         btf__free(obj->btf);
1789
1790         for (i = 0; i < obj->nr_maps; i++) {
1791                 zfree(&obj->maps[i].name);
1792                 if (obj->maps[i].clear_priv)
1793                         obj->maps[i].clear_priv(&obj->maps[i],
1794                                                 obj->maps[i].priv);
1795                 obj->maps[i].priv = NULL;
1796                 obj->maps[i].clear_priv = NULL;
1797         }
1798         zfree(&obj->maps);
1799         obj->nr_maps = 0;
1800
1801         if (obj->programs && obj->nr_programs) {
1802                 for (i = 0; i < obj->nr_programs; i++)
1803                         bpf_program__exit(&obj->programs[i]);
1804         }
1805         zfree(&obj->programs);
1806
1807         list_del(&obj->list);
1808         free(obj);
1809 }
1810
1811 struct bpf_object *
1812 bpf_object__next(struct bpf_object *prev)
1813 {
1814         struct bpf_object *next;
1815
1816         if (!prev)
1817                 next = list_first_entry(&bpf_objects_list,
1818                                         struct bpf_object,
1819                                         list);
1820         else
1821                 next = list_next_entry(prev, list);
1822
1823         /* Empty list is noticed here so don't need checking on entry. */
1824         if (&next->list == &bpf_objects_list)
1825                 return NULL;
1826
1827         return next;
1828 }
1829
1830 const char *bpf_object__name(struct bpf_object *obj)
1831 {
1832         return obj ? obj->path : ERR_PTR(-EINVAL);
1833 }
1834
1835 unsigned int bpf_object__kversion(struct bpf_object *obj)
1836 {
1837         return obj ? obj->kern_version : 0;
1838 }
1839
1840 int bpf_object__btf_fd(const struct bpf_object *obj)
1841 {
1842         return obj->btf ? btf__fd(obj->btf) : -1;
1843 }
1844
1845 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
1846                          bpf_object_clear_priv_t clear_priv)
1847 {
1848         if (obj->priv && obj->clear_priv)
1849                 obj->clear_priv(obj, obj->priv);
1850
1851         obj->priv = priv;
1852         obj->clear_priv = clear_priv;
1853         return 0;
1854 }
1855
1856 void *bpf_object__priv(struct bpf_object *obj)
1857 {
1858         return obj ? obj->priv : ERR_PTR(-EINVAL);
1859 }
1860
1861 struct bpf_program *
1862 bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1863 {
1864         size_t idx;
1865
1866         if (!obj->programs)
1867                 return NULL;
1868         /* First handler */
1869         if (prev == NULL)
1870                 return &obj->programs[0];
1871
1872         if (prev->obj != obj) {
1873                 pr_warning("error: program handler doesn't match object\n");
1874                 return NULL;
1875         }
1876
1877         idx = (prev - obj->programs) + 1;
1878         if (idx >= obj->nr_programs)
1879                 return NULL;
1880         return &obj->programs[idx];
1881 }
1882
1883 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
1884                           bpf_program_clear_priv_t clear_priv)
1885 {
1886         if (prog->priv && prog->clear_priv)
1887                 prog->clear_priv(prog, prog->priv);
1888
1889         prog->priv = priv;
1890         prog->clear_priv = clear_priv;
1891         return 0;
1892 }
1893
1894 void *bpf_program__priv(struct bpf_program *prog)
1895 {
1896         return prog ? prog->priv : ERR_PTR(-EINVAL);
1897 }
1898
1899 const char *bpf_program__title(struct bpf_program *prog, bool needs_copy)
1900 {
1901         const char *title;
1902
1903         title = prog->section_name;
1904         if (needs_copy) {
1905                 title = strdup(title);
1906                 if (!title) {
1907                         pr_warning("failed to strdup program title\n");
1908                         return ERR_PTR(-ENOMEM);
1909                 }
1910         }
1911
1912         return title;
1913 }
1914
1915 int bpf_program__fd(struct bpf_program *prog)
1916 {
1917         return bpf_program__nth_fd(prog, 0);
1918 }
1919
1920 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
1921                           bpf_program_prep_t prep)
1922 {
1923         int *instances_fds;
1924
1925         if (nr_instances <= 0 || !prep)
1926                 return -EINVAL;
1927
1928         if (prog->instances.nr > 0 || prog->instances.fds) {
1929                 pr_warning("Can't set pre-processor after loading\n");
1930                 return -EINVAL;
1931         }
1932
1933         instances_fds = malloc(sizeof(int) * nr_instances);
1934         if (!instances_fds) {
1935                 pr_warning("alloc memory failed for fds\n");
1936                 return -ENOMEM;
1937         }
1938
1939         /* fill all fd with -1 */
1940         memset(instances_fds, -1, sizeof(int) * nr_instances);
1941
1942         prog->instances.nr = nr_instances;
1943         prog->instances.fds = instances_fds;
1944         prog->preprocessor = prep;
1945         return 0;
1946 }
1947
1948 int bpf_program__nth_fd(struct bpf_program *prog, int n)
1949 {
1950         int fd;
1951
1952         if (n >= prog->instances.nr || n < 0) {
1953                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
1954                            n, prog->section_name, prog->instances.nr);
1955                 return -EINVAL;
1956         }
1957
1958         fd = prog->instances.fds[n];
1959         if (fd < 0) {
1960                 pr_warning("%dth instance of program '%s' is invalid\n",
1961                            n, prog->section_name);
1962                 return -ENOENT;
1963         }
1964
1965         return fd;
1966 }
1967
1968 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
1969 {
1970         prog->type = type;
1971 }
1972
1973 static bool bpf_program__is_type(struct bpf_program *prog,
1974                                  enum bpf_prog_type type)
1975 {
1976         return prog ? (prog->type == type) : false;
1977 }
1978
1979 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                   \
1980 int bpf_program__set_##NAME(struct bpf_program *prog)   \
1981 {                                                       \
1982         if (!prog)                                      \
1983                 return -EINVAL;                         \
1984         bpf_program__set_type(prog, TYPE);              \
1985         return 0;                                       \
1986 }                                                       \
1987                                                         \
1988 bool bpf_program__is_##NAME(struct bpf_program *prog)   \
1989 {                                                       \
1990         return bpf_program__is_type(prog, TYPE);        \
1991 }                                                       \
1992
1993 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
1994 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
1995 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
1996 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
1997 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
1998 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
1999 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
2000 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
2001
2002 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
2003                                            enum bpf_attach_type type)
2004 {
2005         prog->expected_attach_type = type;
2006 }
2007
2008 #define BPF_PROG_SEC_FULL(string, ptype, atype) \
2009         { string, sizeof(string) - 1, ptype, atype }
2010
2011 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_FULL(string, ptype, 0)
2012
2013 #define BPF_S_PROG_SEC(string, ptype) \
2014         BPF_PROG_SEC_FULL(string, BPF_PROG_TYPE_CGROUP_SOCK, ptype)
2015
2016 #define BPF_SA_PROG_SEC(string, ptype) \
2017         BPF_PROG_SEC_FULL(string, BPF_PROG_TYPE_CGROUP_SOCK_ADDR, ptype)
2018
2019 static const struct {
2020         const char *sec;
2021         size_t len;
2022         enum bpf_prog_type prog_type;
2023         enum bpf_attach_type expected_attach_type;
2024 } section_names[] = {
2025         BPF_PROG_SEC("socket",          BPF_PROG_TYPE_SOCKET_FILTER),
2026         BPF_PROG_SEC("kprobe/",         BPF_PROG_TYPE_KPROBE),
2027         BPF_PROG_SEC("kretprobe/",      BPF_PROG_TYPE_KPROBE),
2028         BPF_PROG_SEC("classifier",      BPF_PROG_TYPE_SCHED_CLS),
2029         BPF_PROG_SEC("action",          BPF_PROG_TYPE_SCHED_ACT),
2030         BPF_PROG_SEC("tracepoint/",     BPF_PROG_TYPE_TRACEPOINT),
2031         BPF_PROG_SEC("raw_tracepoint/", BPF_PROG_TYPE_RAW_TRACEPOINT),
2032         BPF_PROG_SEC("xdp",             BPF_PROG_TYPE_XDP),
2033         BPF_PROG_SEC("perf_event",      BPF_PROG_TYPE_PERF_EVENT),
2034         BPF_PROG_SEC("cgroup/skb",      BPF_PROG_TYPE_CGROUP_SKB),
2035         BPF_PROG_SEC("cgroup/sock",     BPF_PROG_TYPE_CGROUP_SOCK),
2036         BPF_PROG_SEC("cgroup/dev",      BPF_PROG_TYPE_CGROUP_DEVICE),
2037         BPF_PROG_SEC("lwt_in",          BPF_PROG_TYPE_LWT_IN),
2038         BPF_PROG_SEC("lwt_out",         BPF_PROG_TYPE_LWT_OUT),
2039         BPF_PROG_SEC("lwt_xmit",        BPF_PROG_TYPE_LWT_XMIT),
2040         BPF_PROG_SEC("sockops",         BPF_PROG_TYPE_SOCK_OPS),
2041         BPF_PROG_SEC("sk_skb",          BPF_PROG_TYPE_SK_SKB),
2042         BPF_PROG_SEC("sk_msg",          BPF_PROG_TYPE_SK_MSG),
2043         BPF_SA_PROG_SEC("cgroup/bind4", BPF_CGROUP_INET4_BIND),
2044         BPF_SA_PROG_SEC("cgroup/bind6", BPF_CGROUP_INET6_BIND),
2045         BPF_SA_PROG_SEC("cgroup/connect4", BPF_CGROUP_INET4_CONNECT),
2046         BPF_SA_PROG_SEC("cgroup/connect6", BPF_CGROUP_INET6_CONNECT),
2047         BPF_SA_PROG_SEC("cgroup/sendmsg4", BPF_CGROUP_UDP4_SENDMSG),
2048         BPF_SA_PROG_SEC("cgroup/sendmsg6", BPF_CGROUP_UDP6_SENDMSG),
2049         BPF_S_PROG_SEC("cgroup/post_bind4", BPF_CGROUP_INET4_POST_BIND),
2050         BPF_S_PROG_SEC("cgroup/post_bind6", BPF_CGROUP_INET6_POST_BIND),
2051 };
2052
2053 #undef BPF_PROG_SEC
2054 #undef BPF_PROG_SEC_FULL
2055 #undef BPF_S_PROG_SEC
2056 #undef BPF_SA_PROG_SEC
2057
2058 static int bpf_program__identify_section(struct bpf_program *prog)
2059 {
2060         int i;
2061
2062         if (!prog->section_name)
2063                 goto err;
2064
2065         for (i = 0; i < ARRAY_SIZE(section_names); i++)
2066                 if (strncmp(prog->section_name, section_names[i].sec,
2067                             section_names[i].len) == 0)
2068                         return i;
2069
2070 err:
2071         pr_warning("failed to guess program type based on section name %s\n",
2072                    prog->section_name);
2073
2074         return -1;
2075 }
2076
2077 int bpf_map__fd(struct bpf_map *map)
2078 {
2079         return map ? map->fd : -EINVAL;
2080 }
2081
2082 const struct bpf_map_def *bpf_map__def(struct bpf_map *map)
2083 {
2084         return map ? &map->def : ERR_PTR(-EINVAL);
2085 }
2086
2087 const char *bpf_map__name(struct bpf_map *map)
2088 {
2089         return map ? map->name : NULL;
2090 }
2091
2092 uint32_t bpf_map__btf_key_type_id(const struct bpf_map *map)
2093 {
2094         return map ? map->btf_key_type_id : 0;
2095 }
2096
2097 uint32_t bpf_map__btf_value_type_id(const struct bpf_map *map)
2098 {
2099         return map ? map->btf_value_type_id : 0;
2100 }
2101
2102 int bpf_map__set_priv(struct bpf_map *map, void *priv,
2103                      bpf_map_clear_priv_t clear_priv)
2104 {
2105         if (!map)
2106                 return -EINVAL;
2107
2108         if (map->priv) {
2109                 if (map->clear_priv)
2110                         map->clear_priv(map, map->priv);
2111         }
2112
2113         map->priv = priv;
2114         map->clear_priv = clear_priv;
2115         return 0;
2116 }
2117
2118 void *bpf_map__priv(struct bpf_map *map)
2119 {
2120         return map ? map->priv : ERR_PTR(-EINVAL);
2121 }
2122
2123 struct bpf_map *
2124 bpf_map__next(struct bpf_map *prev, struct bpf_object *obj)
2125 {
2126         size_t idx;
2127         struct bpf_map *s, *e;
2128
2129         if (!obj || !obj->maps)
2130                 return NULL;
2131
2132         s = obj->maps;
2133         e = obj->maps + obj->nr_maps;
2134
2135         if (prev == NULL)
2136                 return s;
2137
2138         if ((prev < s) || (prev >= e)) {
2139                 pr_warning("error in %s: map handler doesn't belong to object\n",
2140                            __func__);
2141                 return NULL;
2142         }
2143
2144         idx = (prev - obj->maps) + 1;
2145         if (idx >= obj->nr_maps)
2146                 return NULL;
2147         return &obj->maps[idx];
2148 }
2149
2150 struct bpf_map *
2151 bpf_object__find_map_by_name(struct bpf_object *obj, const char *name)
2152 {
2153         struct bpf_map *pos;
2154
2155         bpf_map__for_each(pos, obj) {
2156                 if (pos->name && !strcmp(pos->name, name))
2157                         return pos;
2158         }
2159         return NULL;
2160 }
2161
2162 struct bpf_map *
2163 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
2164 {
2165         int i;
2166
2167         for (i = 0; i < obj->nr_maps; i++) {
2168                 if (obj->maps[i].offset == offset)
2169                         return &obj->maps[i];
2170         }
2171         return ERR_PTR(-ENOENT);
2172 }
2173
2174 long libbpf_get_error(const void *ptr)
2175 {
2176         if (IS_ERR(ptr))
2177                 return PTR_ERR(ptr);
2178         return 0;
2179 }
2180
2181 int bpf_prog_load(const char *file, enum bpf_prog_type type,
2182                   struct bpf_object **pobj, int *prog_fd)
2183 {
2184         struct bpf_prog_load_attr attr;
2185
2186         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
2187         attr.file = file;
2188         attr.prog_type = type;
2189         attr.expected_attach_type = 0;
2190
2191         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
2192 }
2193
2194 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
2195                         struct bpf_object **pobj, int *prog_fd)
2196 {
2197         struct bpf_program *prog, *first_prog = NULL;
2198         enum bpf_attach_type expected_attach_type;
2199         enum bpf_prog_type prog_type;
2200         struct bpf_object *obj;
2201         struct bpf_map *map;
2202         int section_idx;
2203         int err;
2204
2205         if (!attr)
2206                 return -EINVAL;
2207         if (!attr->file)
2208                 return -EINVAL;
2209
2210         obj = __bpf_object__open(attr->file, NULL, 0,
2211                                  bpf_prog_type__needs_kver(attr->prog_type));
2212         if (IS_ERR_OR_NULL(obj))
2213                 return -ENOENT;
2214
2215         bpf_object__for_each_program(prog, obj) {
2216                 /*
2217                  * If type is not specified, try to guess it based on
2218                  * section name.
2219                  */
2220                 prog_type = attr->prog_type;
2221                 prog->prog_ifindex = attr->ifindex;
2222                 expected_attach_type = attr->expected_attach_type;
2223                 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
2224                         section_idx = bpf_program__identify_section(prog);
2225                         if (section_idx < 0) {
2226                                 bpf_object__close(obj);
2227                                 return -EINVAL;
2228                         }
2229                         prog_type = section_names[section_idx].prog_type;
2230                         expected_attach_type =
2231                                 section_names[section_idx].expected_attach_type;
2232                 }
2233
2234                 bpf_program__set_type(prog, prog_type);
2235                 bpf_program__set_expected_attach_type(prog,
2236                                                       expected_attach_type);
2237
2238                 if (prog->idx != obj->efile.text_shndx && !first_prog)
2239                         first_prog = prog;
2240         }
2241
2242         bpf_map__for_each(map, obj) {
2243                 map->map_ifindex = attr->ifindex;
2244         }
2245
2246         if (!first_prog) {
2247                 pr_warning("object file doesn't contain bpf program\n");
2248                 bpf_object__close(obj);
2249                 return -ENOENT;
2250         }
2251
2252         err = bpf_object__load(obj);
2253         if (err) {
2254                 bpf_object__close(obj);
2255                 return -EINVAL;
2256         }
2257
2258         *pobj = obj;
2259         *prog_fd = bpf_program__fd(first_prog);
2260         return 0;
2261 }
2262
2263 enum bpf_perf_event_ret
2264 bpf_perf_event_read_simple(void *mem, unsigned long size,
2265                            unsigned long page_size, void **buf, size_t *buf_len,
2266                            bpf_perf_event_print_t fn, void *priv)
2267 {
2268         volatile struct perf_event_mmap_page *header = mem;
2269         __u64 data_tail = header->data_tail;
2270         __u64 data_head = header->data_head;
2271         void *base, *begin, *end;
2272         int ret;
2273
2274         asm volatile("" ::: "memory"); /* in real code it should be smp_rmb() */
2275         if (data_head == data_tail)
2276                 return LIBBPF_PERF_EVENT_CONT;
2277
2278         base = ((char *)header) + page_size;
2279
2280         begin = base + data_tail % size;
2281         end = base + data_head % size;
2282
2283         while (begin != end) {
2284                 struct perf_event_header *ehdr;
2285
2286                 ehdr = begin;
2287                 if (begin + ehdr->size > base + size) {
2288                         long len = base + size - begin;
2289
2290                         if (*buf_len < ehdr->size) {
2291                                 free(*buf);
2292                                 *buf = malloc(ehdr->size);
2293                                 if (!*buf) {
2294                                         ret = LIBBPF_PERF_EVENT_ERROR;
2295                                         break;
2296                                 }
2297                                 *buf_len = ehdr->size;
2298                         }
2299
2300                         memcpy(*buf, begin, len);
2301                         memcpy(*buf + len, base, ehdr->size - len);
2302                         ehdr = (void *)*buf;
2303                         begin = base + ehdr->size - len;
2304                 } else if (begin + ehdr->size == base + size) {
2305                         begin = base;
2306                 } else {
2307                         begin += ehdr->size;
2308                 }
2309
2310                 ret = fn(ehdr, priv);
2311                 if (ret != LIBBPF_PERF_EVENT_CONT)
2312                         break;
2313
2314                 data_tail += ehdr->size;
2315         }
2316
2317         __sync_synchronize(); /* smp_mb() */
2318         header->data_tail = data_tail;
2319
2320         return ret;
2321 }