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         bool has_pseudo_calls;
 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
 405                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 406                      si++) {
 407                         GElf_Sym sym;
 408
 409                         if (!gelf_getsym(symbols, si, &sym))
 410                                 continue;
 411                         if (sym.st_shndx != prog->idx)
 412                                 continue;
 413                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 414                                 continue;
 415
 416                         name = elf_strptr(obj->efile.elf,
 417                                           obj->efile.strtabidx,
 418                                           sym.st_name);
 419                         if (!name) {
 420                                 pr_warning("failed to get sym name string for prog %s\n",
 421                                            prog->section_name);
 422                                 return -LIBBPF_ERRNO__LIBELF;
 423                         }
 424                 }
 425
 426                 if (!name && prog->idx == obj->efile.text_shndx)
 427                         name = ".text";
 428
 429                 if (!name) {
 430                         pr_warning("failed to find sym for prog %s\n",
 431                                    prog->section_name);
 432                         return -EINVAL;
 433                 }
 434
 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                         obj->has_pseudo_calls = true;
 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         struct bpf_map_def *def = &map->def;
1019         const size_t max_name = 256;
1020         int64_t key_size, value_size;
1021         int32_t key_id, value_id;
1022         char name[max_name];
1023
1024         /* Find key type by name from BTF */
1025         if (snprintf(name, max_name, "%s_key", map->name) == max_name) {
1026                 pr_warning("map:%s length of BTF key_type:%s_key is too long\n",
1027                            map->name, map->name);
1028                 return -EINVAL;
1029         }
1030
1031         key_id = btf__find_by_name(btf, name);
1032         if (key_id < 0) {
1033                 pr_debug("map:%s key_type:%s cannot be found in BTF\n",
1034                          map->name, name);
1035                 return key_id;
1036         }
1037
1038         key_size = btf__resolve_size(btf, key_id);
1039         if (key_size < 0) {
1040                 pr_warning("map:%s key_type:%s cannot get the BTF type_size\n",
1041                            map->name, name);
1042                 return key_size;
1043         }
1044
1045         if (def->key_size != key_size) {
1046                 pr_warning("map:%s key_type:%s has BTF type_size:%u != key_size:%u\n",
1047                            map->name, name, (unsigned int)key_size, def->key_size);
1048                 return -EINVAL;
1049         }
1050
1051         /* Find value type from BTF */
1052         if (snprintf(name, max_name, "%s_value", map->name) == max_name) {
1053                 pr_warning("map:%s length of BTF value_type:%s_value is too long\n",
1054                           map->name, map->name);
1055                 return -EINVAL;
1056         }
1057
1058         value_id = btf__find_by_name(btf, name);
1059         if (value_id < 0) {
1060                 pr_debug("map:%s value_type:%s cannot be found in BTF\n",
1061                          map->name, name);
1062                 return value_id;
1063         }
1064
1065         value_size = btf__resolve_size(btf, value_id);
1066         if (value_size < 0) {
1067                 pr_warning("map:%s value_type:%s cannot get the BTF type_size\n",
1068                            map->name, name);
1069                 return value_size;
1070         }
1071
1072         if (def->value_size != value_size) {
1073                 pr_warning("map:%s value_type:%s has BTF type_size:%u != value_size:%u\n",
1074                            map->name, name, (unsigned int)value_size, def->value_size);
1075                 return -EINVAL;
1076         }
1077
1078         map->btf_key_type_id = key_id;
1079         map->btf_value_type_id = value_id;
1080
1081         return 0;
1082 }
1083
1084 static int
1085 bpf_object__create_maps(struct bpf_object *obj)
1086 {
1087         struct bpf_create_map_attr create_attr = {};
1088         unsigned int i;
1089         int err;
1090
1091         for (i = 0; i < obj->nr_maps; i++) {
1092                 struct bpf_map *map = &obj->maps[i];
1093                 struct bpf_map_def *def = &map->def;
1094                 int *pfd = &map->fd;
1095
1096                 create_attr.name = map->name;
1097                 create_attr.map_ifindex = map->map_ifindex;
1098                 create_attr.map_type = def->type;
1099                 create_attr.map_flags = def->map_flags;
1100                 create_attr.key_size = def->key_size;
1101                 create_attr.value_size = def->value_size;
1102                 create_attr.max_entries = def->max_entries;
1103                 create_attr.btf_fd = 0;
1104                 create_attr.btf_key_type_id = 0;
1105                 create_attr.btf_value_type_id = 0;
1106
1107                 if (obj->btf && !bpf_map_find_btf_info(map, obj->btf)) {
1108                         create_attr.btf_fd = btf__fd(obj->btf);
1109                         create_attr.btf_key_type_id = map->btf_key_type_id;
1110                         create_attr.btf_value_type_id = map->btf_value_type_id;
1111                 }
1112
1113                 *pfd = bpf_create_map_xattr(&create_attr);
1114                 if (*pfd < 0 && create_attr.btf_key_type_id) {
1115                         pr_warning("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
1116                                    map->name, strerror(errno), errno);
1117                         create_attr.btf_fd = 0;
1118                         create_attr.btf_key_type_id = 0;
1119                         create_attr.btf_value_type_id = 0;
1120                         map->btf_key_type_id = 0;
1121                         map->btf_value_type_id = 0;
1122                         *pfd = bpf_create_map_xattr(&create_attr);
1123                 }
1124
1125                 if (*pfd < 0) {
1126                         size_t j;
1127
1128                         err = *pfd;
1129                         pr_warning("failed to create map (name: '%s'): %s\n",
1130                                    map->name,
1131                                    strerror(errno));
1132                         for (j = 0; j < i; j++)
1133                                 zclose(obj->maps[j].fd);
1134                         return err;
1135                 }
1136                 pr_debug("create map %s: fd=%d\n", map->name, *pfd);
1137         }
1138
1139         return 0;
1140 }
1141
1142 static int
1143 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
1144                         struct reloc_desc *relo)
1145 {
1146         struct bpf_insn *insn, *new_insn;
1147         struct bpf_program *text;
1148         size_t new_cnt;
1149
1150         if (relo->type != RELO_CALL)
1151                 return -LIBBPF_ERRNO__RELOC;
1152
1153         if (prog->idx == obj->efile.text_shndx) {
1154                 pr_warning("relo in .text insn %d into off %d\n",
1155                            relo->insn_idx, relo->text_off);
1156                 return -LIBBPF_ERRNO__RELOC;
1157         }
1158
1159         if (prog->main_prog_cnt == 0) {
1160                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
1161                 if (!text) {
1162                         pr_warning("no .text section found yet relo into text exist\n");
1163                         return -LIBBPF_ERRNO__RELOC;
1164                 }
1165                 new_cnt = prog->insns_cnt + text->insns_cnt;
1166                 new_insn = realloc(prog->insns, new_cnt * sizeof(*insn));
1167                 if (!new_insn) {
1168                         pr_warning("oom in prog realloc\n");
1169                         return -ENOMEM;
1170                 }
1171                 memcpy(new_insn + prog->insns_cnt, text->insns,
1172                        text->insns_cnt * sizeof(*insn));
1173                 prog->insns = new_insn;
1174                 prog->main_prog_cnt = prog->insns_cnt;
1175                 prog->insns_cnt = new_cnt;
1176                 pr_debug("added %zd insn from %s to prog %s\n",
1177                          text->insns_cnt, text->section_name,
1178                          prog->section_name);
1179         }
1180         insn = &prog->insns[relo->insn_idx];
1181         insn->imm += prog->main_prog_cnt - relo->insn_idx;
1182         return 0;
1183 }
1184
1185 static int
1186 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
1187 {
1188         int i, err;
1189
1190         if (!prog || !prog->reloc_desc)
1191                 return 0;
1192
1193         for (i = 0; i < prog->nr_reloc; i++) {
1194                 if (prog->reloc_desc[i].type == RELO_LD64) {
1195                         struct bpf_insn *insns = prog->insns;
1196                         int insn_idx, map_idx;
1197
1198                         insn_idx = prog->reloc_desc[i].insn_idx;
1199                         map_idx = prog->reloc_desc[i].map_idx;
1200
1201                         if (insn_idx >= (int)prog->insns_cnt) {
1202                                 pr_warning("relocation out of range: '%s'\n",
1203                                            prog->section_name);
1204                                 return -LIBBPF_ERRNO__RELOC;
1205                         }
1206                         insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
1207                         insns[insn_idx].imm = obj->maps[map_idx].fd;
1208                 } else {
1209                         err = bpf_program__reloc_text(prog, obj,
1210                                                       &prog->reloc_desc[i]);
1211                         if (err)
1212                                 return err;
1213                 }
1214         }
1215
1216         zfree(&prog->reloc_desc);
1217         prog->nr_reloc = 0;
1218         return 0;
1219 }
1220
1221
1222 static int
1223 bpf_object__relocate(struct bpf_object *obj)
1224 {
1225         struct bpf_program *prog;
1226         size_t i;
1227         int err;
1228
1229         for (i = 0; i < obj->nr_programs; i++) {
1230                 prog = &obj->programs[i];
1231
1232                 err = bpf_program__relocate(prog, obj);
1233                 if (err) {
1234                         pr_warning("failed to relocate '%s'\n",
1235                                    prog->section_name);
1236                         return err;
1237                 }
1238         }
1239         return 0;
1240 }
1241
1242 static int bpf_object__collect_reloc(struct bpf_object *obj)
1243 {
1244         int i, err;
1245
1246         if (!obj_elf_valid(obj)) {
1247                 pr_warning("Internal error: elf object is closed\n");
1248                 return -LIBBPF_ERRNO__INTERNAL;
1249         }
1250
1251         for (i = 0; i < obj->efile.nr_reloc; i++) {
1252                 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
1253                 Elf_Data *data = obj->efile.reloc[i].data;
1254                 int idx = shdr->sh_info;
1255                 struct bpf_program *prog;
1256
1257                 if (shdr->sh_type != SHT_REL) {
1258                         pr_warning("internal error at %d\n", __LINE__);
1259                         return -LIBBPF_ERRNO__INTERNAL;
1260                 }
1261
1262                 prog = bpf_object__find_prog_by_idx(obj, idx);
1263                 if (!prog) {
1264                         pr_warning("relocation failed: no section(%d)\n", idx);
1265                         return -LIBBPF_ERRNO__RELOC;
1266                 }
1267
1268                 err = bpf_program__collect_reloc(prog,
1269                                                  shdr, data,
1270                                                  obj);
1271                 if (err)
1272                         return err;
1273         }
1274         return 0;
1275 }
1276
1277 static int
1278 load_program(enum bpf_prog_type type, enum bpf_attach_type expected_attach_type,
1279              const char *name, struct bpf_insn *insns, int insns_cnt,
1280              char *license, u32 kern_version, int *pfd, int prog_ifindex)
1281 {
1282         struct bpf_load_program_attr load_attr;
1283         char *log_buf;
1284         int ret;
1285
1286         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
1287         load_attr.prog_type = type;
1288         load_attr.expected_attach_type = expected_attach_type;
1289         load_attr.name = name;
1290         load_attr.insns = insns;
1291         load_attr.insns_cnt = insns_cnt;
1292         load_attr.license = license;
1293         load_attr.kern_version = kern_version;
1294         load_attr.prog_ifindex = prog_ifindex;
1295
1296         if (!load_attr.insns || !load_attr.insns_cnt)
1297                 return -EINVAL;
1298
1299         log_buf = malloc(BPF_LOG_BUF_SIZE);
1300         if (!log_buf)
1301                 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
1302
1303         ret = bpf_load_program_xattr(&load_attr, log_buf, BPF_LOG_BUF_SIZE);
1304
1305         if (ret >= 0) {
1306                 *pfd = ret;
1307                 ret = 0;
1308                 goto out;
1309         }
1310
1311         ret = -LIBBPF_ERRNO__LOAD;
1312         pr_warning("load bpf program failed: %s\n", strerror(errno));
1313
1314         if (log_buf && log_buf[0] != '\0') {
1315                 ret = -LIBBPF_ERRNO__VERIFY;
1316                 pr_warning("-- BEGIN DUMP LOG ---\n");
1317                 pr_warning("\n%s\n", log_buf);
1318                 pr_warning("-- END LOG --\n");
1319         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
1320                 pr_warning("Program too large (%zu insns), at most %d insns\n",
1321                            load_attr.insns_cnt, BPF_MAXINSNS);
1322                 ret = -LIBBPF_ERRNO__PROG2BIG;
1323         } else {
1324                 /* Wrong program type? */
1325                 if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
1326                         int fd;
1327
1328                         load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
1329                         load_attr.expected_attach_type = 0;
1330                         fd = bpf_load_program_xattr(&load_attr, NULL, 0);
1331                         if (fd >= 0) {
1332                                 close(fd);
1333                                 ret = -LIBBPF_ERRNO__PROGTYPE;
1334                                 goto out;
1335                         }
1336                 }
1337
1338                 if (log_buf)
1339                         ret = -LIBBPF_ERRNO__KVER;
1340         }
1341
1342 out:
1343         free(log_buf);
1344         return ret;
1345 }
1346
1347 static int
1348 bpf_program__load(struct bpf_program *prog,
1349                   char *license, u32 kern_version)
1350 {
1351         int err = 0, fd, i;
1352
1353         if (prog->instances.nr < 0 || !prog->instances.fds) {
1354                 if (prog->preprocessor) {
1355                         pr_warning("Internal error: can't load program '%s'\n",
1356                                    prog->section_name);
1357                         return -LIBBPF_ERRNO__INTERNAL;
1358                 }
1359
1360                 prog->instances.fds = malloc(sizeof(int));
1361                 if (!prog->instances.fds) {
1362                         pr_warning("Not enough memory for BPF fds\n");
1363                         return -ENOMEM;
1364                 }
1365                 prog->instances.nr = 1;
1366                 prog->instances.fds[0] = -1;
1367         }
1368
1369         if (!prog->preprocessor) {
1370                 if (prog->instances.nr != 1) {
1371                         pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
1372                                    prog->section_name, prog->instances.nr);
1373                 }
1374                 err = load_program(prog->type, prog->expected_attach_type,
1375                                    prog->name, prog->insns, prog->insns_cnt,
1376                                    license, kern_version, &fd,
1377                                    prog->prog_ifindex);
1378                 if (!err)
1379                         prog->instances.fds[0] = fd;
1380                 goto out;
1381         }
1382
1383         for (i = 0; i < prog->instances.nr; i++) {
1384                 struct bpf_prog_prep_result result;
1385                 bpf_program_prep_t preprocessor = prog->preprocessor;
1386
1387                 bzero(&result, sizeof(result));
1388                 err = preprocessor(prog, i, prog->insns,
1389                                    prog->insns_cnt, &result);
1390                 if (err) {
1391                         pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
1392                                    i, prog->section_name);
1393                         goto out;
1394                 }
1395
1396                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
1397                         pr_debug("Skip loading the %dth instance of program '%s'\n",
1398                                  i, prog->section_name);
1399                         prog->instances.fds[i] = -1;
1400                         if (result.pfd)
1401                                 *result.pfd = -1;
1402                         continue;
1403                 }
1404
1405                 err = load_program(prog->type, prog->expected_attach_type,
1406                                    prog->name, result.new_insn_ptr,
1407                                    result.new_insn_cnt,
1408                                    license, kern_version, &fd,
1409                                    prog->prog_ifindex);
1410
1411                 if (err) {
1412                         pr_warning("Loading the %dth instance of program '%s' failed\n",
1413                                         i, prog->section_name);
1414                         goto out;
1415                 }
1416
1417                 if (result.pfd)
1418                         *result.pfd = fd;
1419                 prog->instances.fds[i] = fd;
1420         }
1421 out:
1422         if (err)
1423                 pr_warning("failed to load program '%s'\n",
1424                            prog->section_name);
1425         zfree(&prog->insns);
1426         prog->insns_cnt = 0;
1427         return err;
1428 }
1429
1430 static bool bpf_program__is_function_storage(struct bpf_program *prog,
1431                                              struct bpf_object *obj)
1432 {
1433         return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
1434 }
1435
1436 static int
1437 bpf_object__load_progs(struct bpf_object *obj)
1438 {
1439         size_t i;
1440         int err;
1441
1442         for (i = 0; i < obj->nr_programs; i++) {
1443                 if (bpf_program__is_function_storage(&obj->programs[i], obj))
1444                         continue;
1445                 err = bpf_program__load(&obj->programs[i],
1446                                         obj->license,
1447                                         obj->kern_version);
1448                 if (err)
1449                         return err;
1450         }
1451         return 0;
1452 }
1453
1454 static bool bpf_prog_type__needs_kver(enum bpf_prog_type type)
1455 {
1456         switch (type) {
1457         case BPF_PROG_TYPE_SOCKET_FILTER:
1458         case BPF_PROG_TYPE_SCHED_CLS:
1459         case BPF_PROG_TYPE_SCHED_ACT:
1460         case BPF_PROG_TYPE_XDP:
1461         case BPF_PROG_TYPE_CGROUP_SKB:
1462         case BPF_PROG_TYPE_CGROUP_SOCK:
1463         case BPF_PROG_TYPE_LWT_IN:
1464         case BPF_PROG_TYPE_LWT_OUT:
1465         case BPF_PROG_TYPE_LWT_XMIT:
1466         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
1467         case BPF_PROG_TYPE_SOCK_OPS:
1468         case BPF_PROG_TYPE_SK_SKB:
1469         case BPF_PROG_TYPE_CGROUP_DEVICE:
1470         case BPF_PROG_TYPE_SK_MSG:
1471         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1472         case BPF_PROG_TYPE_LIRC_MODE2:
1473                 return false;
1474         case BPF_PROG_TYPE_UNSPEC:
1475         case BPF_PROG_TYPE_KPROBE:
1476         case BPF_PROG_TYPE_TRACEPOINT:
1477         case BPF_PROG_TYPE_PERF_EVENT:
1478         case BPF_PROG_TYPE_RAW_TRACEPOINT:
1479         default:
1480                 return true;
1481         }
1482 }
1483
1484 static int bpf_object__validate(struct bpf_object *obj, bool needs_kver)
1485 {
1486         if (needs_kver && obj->kern_version == 0) {
1487                 pr_warning("%s doesn't provide kernel version\n",
1488                            obj->path);
1489                 return -LIBBPF_ERRNO__KVERSION;
1490         }
1491         return 0;
1492 }
1493
1494 static struct bpf_object *
1495 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz,
1496                    bool needs_kver)
1497 {
1498         struct bpf_object *obj;
1499         int err;
1500
1501         if (elf_version(EV_CURRENT) == EV_NONE) {
1502                 pr_warning("failed to init libelf for %s\n", path);
1503                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1504         }
1505
1506         obj = bpf_object__new(path, obj_buf, obj_buf_sz);
1507         if (IS_ERR(obj))
1508                 return obj;
1509
1510         CHECK_ERR(bpf_object__elf_init(obj), err, out);
1511         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
1512         CHECK_ERR(bpf_object__elf_collect(obj), err, out);
1513         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
1514         CHECK_ERR(bpf_object__validate(obj, needs_kver), err, out);
1515
1516         bpf_object__elf_finish(obj);
1517         return obj;
1518 out:
1519         bpf_object__close(obj);
1520         return ERR_PTR(err);
1521 }
1522
1523 struct bpf_object *bpf_object__open(const char *path)
1524 {
1525         /* param validation */
1526         if (!path)
1527                 return NULL;
1528
1529         pr_debug("loading %s\n", path);
1530
1531         return __bpf_object__open(path, NULL, 0, true);
1532 }
1533
1534 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
1535                                            size_t obj_buf_sz,
1536                                            const char *name)
1537 {
1538         char tmp_name[64];
1539
1540         /* param validation */
1541         if (!obj_buf || obj_buf_sz <= 0)
1542                 return NULL;
1543
1544         if (!name) {
1545                 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
1546                          (unsigned long)obj_buf,
1547                          (unsigned long)obj_buf_sz);
1548                 tmp_name[sizeof(tmp_name) - 1] = '\0';
1549                 name = tmp_name;
1550         }
1551         pr_debug("loading object '%s' from buffer\n",
1552                  name);
1553
1554         return __bpf_object__open(name, obj_buf, obj_buf_sz, true);
1555 }
1556
1557 int bpf_object__unload(struct bpf_object *obj)
1558 {
1559         size_t i;
1560
1561         if (!obj)
1562                 return -EINVAL;
1563
1564         for (i = 0; i < obj->nr_maps; i++)
1565                 zclose(obj->maps[i].fd);
1566
1567         for (i = 0; i < obj->nr_programs; i++)
1568                 bpf_program__unload(&obj->programs[i]);
1569
1570         return 0;
1571 }
1572
1573 int bpf_object__load(struct bpf_object *obj)
1574 {
1575         int err;
1576
1577         if (!obj)
1578                 return -EINVAL;
1579
1580         if (obj->loaded) {
1581                 pr_warning("object should not be loaded twice\n");
1582                 return -EINVAL;
1583         }
1584
1585         obj->loaded = true;
1586
1587         CHECK_ERR(bpf_object__create_maps(obj), err, out);
1588         CHECK_ERR(bpf_object__relocate(obj), err, out);
1589         CHECK_ERR(bpf_object__load_progs(obj), err, out);
1590
1591         return 0;
1592 out:
1593         bpf_object__unload(obj);
1594         pr_warning("failed to load object '%s'\n", obj->path);
1595         return err;
1596 }
1597
1598 static int check_path(const char *path)
1599 {
1600         struct statfs st_fs;
1601         char *dname, *dir;
1602         int err = 0;
1603
1604         if (path == NULL)
1605                 return -EINVAL;
1606
1607         dname = strdup(path);
1608         if (dname == NULL)
1609                 return -ENOMEM;
1610
1611         dir = dirname(dname);
1612         if (statfs(dir, &st_fs)) {
1613                 pr_warning("failed to statfs %s: %s\n", dir, strerror(errno));
1614                 err = -errno;
1615         }
1616         free(dname);
1617
1618         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
1619                 pr_warning("specified path %s is not on BPF FS\n", path);
1620                 err = -EINVAL;
1621         }
1622
1623         return err;
1624 }
1625
1626 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
1627                               int instance)
1628 {
1629         int err;
1630
1631         err = check_path(path);
1632         if (err)
1633                 return err;
1634
1635         if (prog == NULL) {
1636                 pr_warning("invalid program pointer\n");
1637                 return -EINVAL;
1638         }
1639
1640         if (instance < 0 || instance >= prog->instances.nr) {
1641                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
1642                            instance, prog->section_name, prog->instances.nr);
1643                 return -EINVAL;
1644         }
1645
1646         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
1647                 pr_warning("failed to pin program: %s\n", strerror(errno));
1648                 return -errno;
1649         }
1650         pr_debug("pinned program '%s'\n", path);
1651
1652         return 0;
1653 }
1654
1655 static int make_dir(const char *path)
1656 {
1657         int err = 0;
1658
1659         if (mkdir(path, 0700) && errno != EEXIST)
1660                 err = -errno;
1661
1662         if (err)
1663                 pr_warning("failed to mkdir %s: %s\n", path, strerror(-err));
1664         return err;
1665 }
1666
1667 int bpf_program__pin(struct bpf_program *prog, const char *path)
1668 {
1669         int i, err;
1670
1671         err = check_path(path);
1672         if (err)
1673                 return err;
1674
1675         if (prog == NULL) {
1676                 pr_warning("invalid program pointer\n");
1677                 return -EINVAL;
1678         }
1679
1680         if (prog->instances.nr <= 0) {
1681                 pr_warning("no instances of prog %s to pin\n",
1682                            prog->section_name);
1683                 return -EINVAL;
1684         }
1685
1686         err = make_dir(path);
1687         if (err)
1688                 return err;
1689
1690         for (i = 0; i < prog->instances.nr; i++) {
1691                 char buf[PATH_MAX];
1692                 int len;
1693
1694                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
1695                 if (len < 0)
1696                         return -EINVAL;
1697                 else if (len >= PATH_MAX)
1698                         return -ENAMETOOLONG;
1699
1700                 err = bpf_program__pin_instance(prog, buf, i);
1701                 if (err)
1702                         return err;
1703         }
1704
1705         return 0;
1706 }
1707
1708 int bpf_map__pin(struct bpf_map *map, const char *path)
1709 {
1710         int err;
1711
1712         err = check_path(path);
1713         if (err)
1714                 return err;
1715
1716         if (map == NULL) {
1717                 pr_warning("invalid map pointer\n");
1718                 return -EINVAL;
1719         }
1720
1721         if (bpf_obj_pin(map->fd, path)) {
1722                 pr_warning("failed to pin map: %s\n", strerror(errno));
1723                 return -errno;
1724         }
1725
1726         pr_debug("pinned map '%s'\n", path);
1727         return 0;
1728 }
1729
1730 int bpf_object__pin(struct bpf_object *obj, const char *path)
1731 {
1732         struct bpf_program *prog;
1733         struct bpf_map *map;
1734         int err;
1735
1736         if (!obj)
1737                 return -ENOENT;
1738
1739         if (!obj->loaded) {
1740                 pr_warning("object not yet loaded; load it first\n");
1741                 return -ENOENT;
1742         }
1743
1744         err = make_dir(path);
1745         if (err)
1746                 return err;
1747
1748         bpf_map__for_each(map, obj) {
1749                 char buf[PATH_MAX];
1750                 int len;
1751
1752                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1753                                bpf_map__name(map));
1754                 if (len < 0)
1755                         return -EINVAL;
1756                 else if (len >= PATH_MAX)
1757                         return -ENAMETOOLONG;
1758
1759                 err = bpf_map__pin(map, buf);
1760                 if (err)
1761                         return err;
1762         }
1763
1764         bpf_object__for_each_program(prog, obj) {
1765                 char buf[PATH_MAX];
1766                 int len;
1767
1768                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1769                                prog->section_name);
1770                 if (len < 0)
1771                         return -EINVAL;
1772                 else if (len >= PATH_MAX)
1773                         return -ENAMETOOLONG;
1774
1775                 err = bpf_program__pin(prog, buf);
1776                 if (err)
1777                         return err;
1778         }
1779
1780         return 0;
1781 }
1782
1783 void bpf_object__close(struct bpf_object *obj)
1784 {
1785         size_t i;
1786
1787         if (!obj)
1788                 return;
1789
1790         if (obj->clear_priv)
1791                 obj->clear_priv(obj, obj->priv);
1792
1793         bpf_object__elf_finish(obj);
1794         bpf_object__unload(obj);
1795         btf__free(obj->btf);
1796
1797         for (i = 0; i < obj->nr_maps; i++) {
1798                 zfree(&obj->maps[i].name);
1799                 if (obj->maps[i].clear_priv)
1800                         obj->maps[i].clear_priv(&obj->maps[i],
1801                                                 obj->maps[i].priv);
1802                 obj->maps[i].priv = NULL;
1803                 obj->maps[i].clear_priv = NULL;
1804         }
1805         zfree(&obj->maps);
1806         obj->nr_maps = 0;
1807
1808         if (obj->programs && obj->nr_programs) {
1809                 for (i = 0; i < obj->nr_programs; i++)
1810                         bpf_program__exit(&obj->programs[i]);
1811         }
1812         zfree(&obj->programs);
1813
1814         list_del(&obj->list);
1815         free(obj);
1816 }
1817
1818 struct bpf_object *
1819 bpf_object__next(struct bpf_object *prev)
1820 {
1821         struct bpf_object *next;
1822
1823         if (!prev)
1824                 next = list_first_entry(&bpf_objects_list,
1825                                         struct bpf_object,
1826                                         list);
1827         else
1828                 next = list_next_entry(prev, list);
1829
1830         /* Empty list is noticed here so don't need checking on entry. */
1831         if (&next->list == &bpf_objects_list)
1832                 return NULL;
1833
1834         return next;
1835 }
1836
1837 const char *bpf_object__name(struct bpf_object *obj)
1838 {
1839         return obj ? obj->path : ERR_PTR(-EINVAL);
1840 }
1841
1842 unsigned int bpf_object__kversion(struct bpf_object *obj)
1843 {
1844         return obj ? obj->kern_version : 0;
1845 }
1846
1847 int bpf_object__btf_fd(const struct bpf_object *obj)
1848 {
1849         return obj->btf ? btf__fd(obj->btf) : -1;
1850 }
1851
1852 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
1853                          bpf_object_clear_priv_t clear_priv)
1854 {
1855         if (obj->priv && obj->clear_priv)
1856                 obj->clear_priv(obj, obj->priv);
1857
1858         obj->priv = priv;
1859         obj->clear_priv = clear_priv;
1860         return 0;
1861 }
1862
1863 void *bpf_object__priv(struct bpf_object *obj)
1864 {
1865         return obj ? obj->priv : ERR_PTR(-EINVAL);
1866 }
1867
1868 static struct bpf_program *
1869 __bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1870 {
1871         size_t idx;
1872
1873         if (!obj->programs)
1874                 return NULL;
1875         /* First handler */
1876         if (prev == NULL)
1877                 return &obj->programs[0];
1878
1879         if (prev->obj != obj) {
1880                 pr_warning("error: program handler doesn't match object\n");
1881                 return NULL;
1882         }
1883
1884         idx = (prev - obj->programs) + 1;
1885         if (idx >= obj->nr_programs)
1886                 return NULL;
1887         return &obj->programs[idx];
1888 }
1889
1890 struct bpf_program *
1891 bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1892 {
1893         struct bpf_program *prog = prev;
1894
1895         do {
1896                 prog = __bpf_program__next(prog, obj);
1897         } while (prog && bpf_program__is_function_storage(prog, obj));
1898
1899         return prog;
1900 }
1901
1902 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
1903                           bpf_program_clear_priv_t clear_priv)
1904 {
1905         if (prog->priv && prog->clear_priv)
1906                 prog->clear_priv(prog, prog->priv);
1907
1908         prog->priv = priv;
1909         prog->clear_priv = clear_priv;
1910         return 0;
1911 }
1912
1913 void *bpf_program__priv(struct bpf_program *prog)
1914 {
1915         return prog ? prog->priv : ERR_PTR(-EINVAL);
1916 }
1917
1918 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
1919 {
1920         prog->prog_ifindex = ifindex;
1921 }
1922
1923 const char *bpf_program__title(struct bpf_program *prog, bool needs_copy)
1924 {
1925         const char *title;
1926
1927         title = prog->section_name;
1928         if (needs_copy) {
1929                 title = strdup(title);
1930                 if (!title) {
1931                         pr_warning("failed to strdup program title\n");
1932                         return ERR_PTR(-ENOMEM);
1933                 }
1934         }
1935
1936         return title;
1937 }
1938
1939 int bpf_program__fd(struct bpf_program *prog)
1940 {
1941         return bpf_program__nth_fd(prog, 0);
1942 }
1943
1944 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
1945                           bpf_program_prep_t prep)
1946 {
1947         int *instances_fds;
1948
1949         if (nr_instances <= 0 || !prep)
1950                 return -EINVAL;
1951
1952         if (prog->instances.nr > 0 || prog->instances.fds) {
1953                 pr_warning("Can't set pre-processor after loading\n");
1954                 return -EINVAL;
1955         }
1956
1957         instances_fds = malloc(sizeof(int) * nr_instances);
1958         if (!instances_fds) {
1959                 pr_warning("alloc memory failed for fds\n");
1960                 return -ENOMEM;
1961         }
1962
1963         /* fill all fd with -1 */
1964         memset(instances_fds, -1, sizeof(int) * nr_instances);
1965
1966         prog->instances.nr = nr_instances;
1967         prog->instances.fds = instances_fds;
1968         prog->preprocessor = prep;
1969         return 0;
1970 }
1971
1972 int bpf_program__nth_fd(struct bpf_program *prog, int n)
1973 {
1974         int fd;
1975
1976         if (n >= prog->instances.nr || n < 0) {
1977                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
1978                            n, prog->section_name, prog->instances.nr);
1979                 return -EINVAL;
1980         }
1981
1982         fd = prog->instances.fds[n];
1983         if (fd < 0) {
1984                 pr_warning("%dth instance of program '%s' is invalid\n",
1985                            n, prog->section_name);
1986                 return -ENOENT;
1987         }
1988
1989         return fd;
1990 }
1991
1992 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
1993 {
1994         prog->type = type;
1995 }
1996
1997 static bool bpf_program__is_type(struct bpf_program *prog,
1998                                  enum bpf_prog_type type)
1999 {
2000         return prog ? (prog->type == type) : false;
2001 }
2002
2003 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                   \
2004 int bpf_program__set_##NAME(struct bpf_program *prog)   \
2005 {                                                       \
2006         if (!prog)                                      \
2007                 return -EINVAL;                         \
2008         bpf_program__set_type(prog, TYPE);              \
2009         return 0;                                       \
2010 }                                                       \
2011                                                         \
2012 bool bpf_program__is_##NAME(struct bpf_program *prog)   \
2013 {                                                       \
2014         return bpf_program__is_type(prog, TYPE);        \
2015 }                                                       \
2016
2017 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
2018 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
2019 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
2020 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
2021 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
2022 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
2023 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
2024 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
2025
2026 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
2027                                            enum bpf_attach_type type)
2028 {
2029         prog->expected_attach_type = type;
2030 }
2031
2032 #define BPF_PROG_SEC_FULL(string, ptype, atype) \
2033         { string, sizeof(string) - 1, ptype, atype }
2034
2035 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_FULL(string, ptype, 0)
2036
2037 #define BPF_S_PROG_SEC(string, ptype) \
2038         BPF_PROG_SEC_FULL(string, BPF_PROG_TYPE_CGROUP_SOCK, ptype)
2039
2040 #define BPF_SA_PROG_SEC(string, ptype) \
2041         BPF_PROG_SEC_FULL(string, BPF_PROG_TYPE_CGROUP_SOCK_ADDR, ptype)
2042
2043 static const struct {
2044         const char *sec;
2045         size_t len;
2046         enum bpf_prog_type prog_type;
2047         enum bpf_attach_type expected_attach_type;
2048 } section_names[] = {
2049         BPF_PROG_SEC("socket",          BPF_PROG_TYPE_SOCKET_FILTER),
2050         BPF_PROG_SEC("kprobe/",         BPF_PROG_TYPE_KPROBE),
2051         BPF_PROG_SEC("kretprobe/",      BPF_PROG_TYPE_KPROBE),
2052         BPF_PROG_SEC("classifier",      BPF_PROG_TYPE_SCHED_CLS),
2053         BPF_PROG_SEC("action",          BPF_PROG_TYPE_SCHED_ACT),
2054         BPF_PROG_SEC("tracepoint/",     BPF_PROG_TYPE_TRACEPOINT),
2055         BPF_PROG_SEC("raw_tracepoint/", BPF_PROG_TYPE_RAW_TRACEPOINT),
2056         BPF_PROG_SEC("xdp",             BPF_PROG_TYPE_XDP),
2057         BPF_PROG_SEC("perf_event",      BPF_PROG_TYPE_PERF_EVENT),
2058         BPF_PROG_SEC("cgroup/skb",      BPF_PROG_TYPE_CGROUP_SKB),
2059         BPF_PROG_SEC("cgroup/sock",     BPF_PROG_TYPE_CGROUP_SOCK),
2060         BPF_PROG_SEC("cgroup/dev",      BPF_PROG_TYPE_CGROUP_DEVICE),
2061         BPF_PROG_SEC("lwt_in",          BPF_PROG_TYPE_LWT_IN),
2062         BPF_PROG_SEC("lwt_out",         BPF_PROG_TYPE_LWT_OUT),
2063         BPF_PROG_SEC("lwt_xmit",        BPF_PROG_TYPE_LWT_XMIT),
2064         BPF_PROG_SEC("lwt_seg6local",   BPF_PROG_TYPE_LWT_SEG6LOCAL),
2065         BPF_PROG_SEC("sockops",         BPF_PROG_TYPE_SOCK_OPS),
2066         BPF_PROG_SEC("sk_skb",          BPF_PROG_TYPE_SK_SKB),
2067         BPF_PROG_SEC("sk_msg",          BPF_PROG_TYPE_SK_MSG),
2068         BPF_PROG_SEC("lirc_mode2",      BPF_PROG_TYPE_LIRC_MODE2),
2069         BPF_SA_PROG_SEC("cgroup/bind4", BPF_CGROUP_INET4_BIND),
2070         BPF_SA_PROG_SEC("cgroup/bind6", BPF_CGROUP_INET6_BIND),
2071         BPF_SA_PROG_SEC("cgroup/connect4", BPF_CGROUP_INET4_CONNECT),
2072         BPF_SA_PROG_SEC("cgroup/connect6", BPF_CGROUP_INET6_CONNECT),
2073         BPF_SA_PROG_SEC("cgroup/sendmsg4", BPF_CGROUP_UDP4_SENDMSG),
2074         BPF_SA_PROG_SEC("cgroup/sendmsg6", BPF_CGROUP_UDP6_SENDMSG),
2075         BPF_S_PROG_SEC("cgroup/post_bind4", BPF_CGROUP_INET4_POST_BIND),
2076         BPF_S_PROG_SEC("cgroup/post_bind6", BPF_CGROUP_INET6_POST_BIND),
2077 };
2078
2079 #undef BPF_PROG_SEC
2080 #undef BPF_PROG_SEC_FULL
2081 #undef BPF_S_PROG_SEC
2082 #undef BPF_SA_PROG_SEC
2083
2084 static int bpf_program__identify_section(struct bpf_program *prog)
2085 {
2086         int i;
2087
2088         if (!prog->section_name)
2089                 goto err;
2090
2091         for (i = 0; i < ARRAY_SIZE(section_names); i++)
2092                 if (strncmp(prog->section_name, section_names[i].sec,
2093                             section_names[i].len) == 0)
2094                         return i;
2095
2096 err:
2097         pr_warning("failed to guess program type based on section name %s\n",
2098                    prog->section_name);
2099
2100         return -1;
2101 }
2102
2103 int bpf_map__fd(struct bpf_map *map)
2104 {
2105         return map ? map->fd : -EINVAL;
2106 }
2107
2108 const struct bpf_map_def *bpf_map__def(struct bpf_map *map)
2109 {
2110         return map ? &map->def : ERR_PTR(-EINVAL);
2111 }
2112
2113 const char *bpf_map__name(struct bpf_map *map)
2114 {
2115         return map ? map->name : NULL;
2116 }
2117
2118 uint32_t bpf_map__btf_key_type_id(const struct bpf_map *map)
2119 {
2120         return map ? map->btf_key_type_id : 0;
2121 }
2122
2123 uint32_t bpf_map__btf_value_type_id(const struct bpf_map *map)
2124 {
2125         return map ? map->btf_value_type_id : 0;
2126 }
2127
2128 int bpf_map__set_priv(struct bpf_map *map, void *priv,
2129                      bpf_map_clear_priv_t clear_priv)
2130 {
2131         if (!map)
2132                 return -EINVAL;
2133
2134         if (map->priv) {
2135                 if (map->clear_priv)
2136                         map->clear_priv(map, map->priv);
2137         }
2138
2139         map->priv = priv;
2140         map->clear_priv = clear_priv;
2141         return 0;
2142 }
2143
2144 void *bpf_map__priv(struct bpf_map *map)
2145 {
2146         return map ? map->priv : ERR_PTR(-EINVAL);
2147 }
2148
2149 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
2150 {
2151         map->map_ifindex = ifindex;
2152 }
2153
2154 struct bpf_map *
2155 bpf_map__next(struct bpf_map *prev, struct bpf_object *obj)
2156 {
2157         size_t idx;
2158         struct bpf_map *s, *e;
2159
2160         if (!obj || !obj->maps)
2161                 return NULL;
2162
2163         s = obj->maps;
2164         e = obj->maps + obj->nr_maps;
2165
2166         if (prev == NULL)
2167                 return s;
2168
2169         if ((prev < s) || (prev >= e)) {
2170                 pr_warning("error in %s: map handler doesn't belong to object\n",
2171                            __func__);
2172                 return NULL;
2173         }
2174
2175         idx = (prev - obj->maps) + 1;
2176         if (idx >= obj->nr_maps)
2177                 return NULL;
2178         return &obj->maps[idx];
2179 }
2180
2181 struct bpf_map *
2182 bpf_object__find_map_by_name(struct bpf_object *obj, const char *name)
2183 {
2184         struct bpf_map *pos;
2185
2186         bpf_map__for_each(pos, obj) {
2187                 if (pos->name && !strcmp(pos->name, name))
2188                         return pos;
2189         }
2190         return NULL;
2191 }
2192
2193 struct bpf_map *
2194 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
2195 {
2196         int i;
2197
2198         for (i = 0; i < obj->nr_maps; i++) {
2199                 if (obj->maps[i].offset == offset)
2200                         return &obj->maps[i];
2201         }
2202         return ERR_PTR(-ENOENT);
2203 }
2204
2205 long libbpf_get_error(const void *ptr)
2206 {
2207         if (IS_ERR(ptr))
2208                 return PTR_ERR(ptr);
2209         return 0;
2210 }
2211
2212 int bpf_prog_load(const char *file, enum bpf_prog_type type,
2213                   struct bpf_object **pobj, int *prog_fd)
2214 {
2215         struct bpf_prog_load_attr attr;
2216
2217         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
2218         attr.file = file;
2219         attr.prog_type = type;
2220         attr.expected_attach_type = 0;
2221
2222         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
2223 }
2224
2225 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
2226                         struct bpf_object **pobj, int *prog_fd)
2227 {
2228         struct bpf_program *prog, *first_prog = NULL;
2229         enum bpf_attach_type expected_attach_type;
2230         enum bpf_prog_type prog_type;
2231         struct bpf_object *obj;
2232         struct bpf_map *map;
2233         int section_idx;
2234         int err;
2235
2236         if (!attr)
2237                 return -EINVAL;
2238         if (!attr->file)
2239                 return -EINVAL;
2240
2241         obj = __bpf_object__open(attr->file, NULL, 0,
2242                                  bpf_prog_type__needs_kver(attr->prog_type));
2243         if (IS_ERR_OR_NULL(obj))
2244                 return -ENOENT;
2245
2246         bpf_object__for_each_program(prog, obj) {
2247                 /*
2248                  * If type is not specified, try to guess it based on
2249                  * section name.
2250                  */
2251                 prog_type = attr->prog_type;
2252                 prog->prog_ifindex = attr->ifindex;
2253                 expected_attach_type = attr->expected_attach_type;
2254                 if (prog_type == BPF_PROG_TYPE_UNSPEC) {
2255                         section_idx = bpf_program__identify_section(prog);
2256                         if (section_idx < 0) {
2257                                 bpf_object__close(obj);
2258                                 return -EINVAL;
2259                         }
2260                         prog_type = section_names[section_idx].prog_type;
2261                         expected_attach_type =
2262                                 section_names[section_idx].expected_attach_type;
2263                 }
2264
2265                 bpf_program__set_type(prog, prog_type);
2266                 bpf_program__set_expected_attach_type(prog,
2267                                                       expected_attach_type);
2268
2269                 if (!bpf_program__is_function_storage(prog, obj) && !first_prog)
2270                         first_prog = prog;
2271         }
2272
2273         bpf_map__for_each(map, obj) {
2274                 map->map_ifindex = attr->ifindex;
2275         }
2276
2277         if (!first_prog) {
2278                 pr_warning("object file doesn't contain bpf program\n");
2279                 bpf_object__close(obj);
2280                 return -ENOENT;
2281         }
2282
2283         err = bpf_object__load(obj);
2284         if (err) {
2285                 bpf_object__close(obj);
2286                 return -EINVAL;
2287         }
2288
2289         *pobj = obj;
2290         *prog_fd = bpf_program__fd(first_prog);
2291         return 0;
2292 }
2293
2294 enum bpf_perf_event_ret
2295 bpf_perf_event_read_simple(void *mem, unsigned long size,
2296                            unsigned long page_size, void **buf, size_t *buf_len,
2297                            bpf_perf_event_print_t fn, void *priv)
2298 {
2299         volatile struct perf_event_mmap_page *header = mem;
2300         __u64 data_tail = header->data_tail;
2301         __u64 data_head = header->data_head;
2302         void *base, *begin, *end;
2303         int ret;
2304
2305         asm volatile("" ::: "memory"); /* in real code it should be smp_rmb() */
2306         if (data_head == data_tail)
2307                 return LIBBPF_PERF_EVENT_CONT;
2308
2309         base = ((char *)header) + page_size;
2310
2311         begin = base + data_tail % size;
2312         end = base + data_head % size;
2313
2314         while (begin != end) {
2315                 struct perf_event_header *ehdr;
2316
2317                 ehdr = begin;
2318                 if (begin + ehdr->size > base + size) {
2319                         long len = base + size - begin;
2320
2321                         if (*buf_len < ehdr->size) {
2322                                 free(*buf);
2323                                 *buf = malloc(ehdr->size);
2324                                 if (!*buf) {
2325                                         ret = LIBBPF_PERF_EVENT_ERROR;
2326                                         break;
2327                                 }
2328                                 *buf_len = ehdr->size;
2329                         }
2330
2331                         memcpy(*buf, begin, len);
2332                         memcpy(*buf + len, base, ehdr->size - len);
2333                         ehdr = (void *)*buf;
2334                         begin = base + ehdr->size - len;
2335                 } else if (begin + ehdr->size == base + size) {
2336                         begin = base;
2337                 } else {
2338                         begin += ehdr->size;
2339                 }
2340
2341                 ret = fn(ehdr, priv);
2342                 if (ret != LIBBPF_PERF_EVENT_CONT)
2343                         break;
2344
2345                 data_tail += ehdr->size;
2346         }
2347
2348         __sync_synchronize(); /* smp_mb() */
2349         header->data_tail = data_tail;
2350
2351         return ret;
2352 }