tools/lib/bpf/libbpf.c

   1 /*
   2  * Common eBPF ELF object loading operations.
   3  *
   4  * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org>
   5  * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
   6  * Copyright (C) 2015 Huawei Inc.
   7  */
   8
   9 #include <stdlib.h>
  10 #include <stdio.h>
  11 #include <stdarg.h>
  12 #include <inttypes.h>
  13 #include <string.h>
  14 #include <unistd.h>
  15 #include <fcntl.h>
  16 #include <errno.h>
  17 #include <asm/unistd.h>
  18 #include <linux/kernel.h>
  19 #include <linux/bpf.h>
  20 #include <linux/list.h>
  21 #include <libelf.h>
  22 #include <gelf.h>
  23
  24 #include "libbpf.h"
  25 #include "bpf.h"
  26
  27 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  28
  29 __printf(1, 2)
  30 static int __base_pr(const char *format, ...)
  31 {
  32         va_list args;
  33         int err;
  34
  35         va_start(args, format);
  36         err = vfprintf(stderr, format, args);
  37         va_end(args);
  38         return err;
  39 }
  40
  41 static __printf(1, 2) libbpf_print_fn_t __pr_warning = __base_pr;
  42 static __printf(1, 2) libbpf_print_fn_t __pr_info = __base_pr;
  43 static __printf(1, 2) libbpf_print_fn_t __pr_debug;
  44
  45 #define __pr(func, fmt, ...)    \
  46 do {                            \
  47         if ((func))             \
  48                 (func)("libbpf: " fmt, ##__VA_ARGS__); \
  49 } while (0)
  50
  51 #define pr_warning(fmt, ...)    __pr(__pr_warning, fmt, ##__VA_ARGS__)
  52 #define pr_info(fmt, ...)       __pr(__pr_info, fmt, ##__VA_ARGS__)
  53 #define pr_debug(fmt, ...)      __pr(__pr_debug, fmt, ##__VA_ARGS__)
  54
  55 void libbpf_set_print(libbpf_print_fn_t warn,
  56                       libbpf_print_fn_t info,
  57                       libbpf_print_fn_t debug)
  58 {
  59         __pr_warning = warn;
  60         __pr_info = info;
  61         __pr_debug = debug;
  62 }
  63
  64 #define STRERR_BUFSIZE  128
  65
  66 #define ERRNO_OFFSET(e)         ((e) - __LIBBPF_ERRNO__START)
  67 #define ERRCODE_OFFSET(c)       ERRNO_OFFSET(LIBBPF_ERRNO__##c)
  68 #define NR_ERRNO        (__LIBBPF_ERRNO__END - __LIBBPF_ERRNO__START)
  69
  70 static const char *libbpf_strerror_table[NR_ERRNO] = {
  71         [ERRCODE_OFFSET(LIBELF)]        = "Something wrong in libelf",
  72         [ERRCODE_OFFSET(FORMAT)]        = "BPF object format invalid",
  73         [ERRCODE_OFFSET(KVERSION)]      = "'version' section incorrect or lost",
  74         [ERRCODE_OFFSET(ENDIAN)]        = "Endian missmatch",
  75         [ERRCODE_OFFSET(INTERNAL)]      = "Internal error in libbpf",
  76         [ERRCODE_OFFSET(RELOC)]         = "Relocation failed",
  77         [ERRCODE_OFFSET(VERIFY)]        = "Kernel verifier blocks program loading",
  78         [ERRCODE_OFFSET(PROG2BIG)]      = "Program too big",
  79         [ERRCODE_OFFSET(KVER)]          = "Incorrect kernel version",
  80 };
  81
  82 int libbpf_strerror(int err, char *buf, size_t size)
  83 {
  84         if (!buf || !size)
  85                 return -1;
  86
  87         err = err > 0 ? err : -err;
  88
  89         if (err < __LIBBPF_ERRNO__START) {
  90                 int ret;
  91
  92                 ret = strerror_r(err, buf, size);
  93                 buf[size - 1] = '\0';
  94                 return ret;
  95         }
  96
  97         if (err < __LIBBPF_ERRNO__END) {
  98                 const char *msg;
  99
 100                 msg = libbpf_strerror_table[ERRNO_OFFSET(err)];
 101                 snprintf(buf, size, "%s", msg);
 102                 buf[size - 1] = '\0';
 103                 return 0;
 104         }
 105
 106         snprintf(buf, size, "Unknown libbpf error %d", err);
 107         buf[size - 1] = '\0';
 108         return -1;
 109 }
 110
 111 #define CHECK_ERR(action, err, out) do {        \
 112         err = action;                   \
 113         if (err)                        \
 114                 goto out;               \
 115 } while(0)
 116
 117
 118 /* Copied from tools/perf/util/util.h */
 119 #ifndef zfree
 120 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 121 #endif
 122
 123 #ifndef zclose
 124 # define zclose(fd) ({                  \
 125         int ___err = 0;                 \
 126         if ((fd) >= 0)                  \
 127                 ___err = close((fd));   \
 128         fd = -1;                        \
 129         ___err; })
 130 #endif
 131
 132 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 133 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 134 #else
 135 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 136 #endif
 137
 138 /*
 139  * bpf_prog should be a better name but it has been used in
 140  * linux/filter.h.
 141  */
 142 struct bpf_program {
 143         /* Index in elf obj file, for relocation use. */
 144         int idx;
 145         char *section_name;
 146         struct bpf_insn *insns;
 147         size_t insns_cnt;
 148
 149         struct {
 150                 int insn_idx;
 151                 int map_idx;
 152         } *reloc_desc;
 153         int nr_reloc;
 154
 155         struct {
 156                 int nr;
 157                 int *fds;
 158         } instances;
 159         bpf_program_prep_t preprocessor;
 160
 161         struct bpf_object *obj;
 162         void *priv;
 163         bpf_program_clear_priv_t clear_priv;
 164 };
 165
 166 struct bpf_map {
 167         int fd;
 168         char *name;
 169         struct bpf_map_def def;
 170         void *priv;
 171         bpf_map_clear_priv_t clear_priv;
 172 };
 173
 174 static LIST_HEAD(bpf_objects_list);
 175
 176 struct bpf_object {
 177         char license[64];
 178         u32 kern_version;
 179
 180         struct bpf_program *programs;
 181         size_t nr_programs;
 182         struct bpf_map *maps;
 183         size_t nr_maps;
 184
 185         bool loaded;
 186
 187         /*
 188          * Information when doing elf related work. Only valid if fd
 189          * is valid.
 190          */
 191         struct {
 192                 int fd;
 193                 void *obj_buf;
 194                 size_t obj_buf_sz;
 195                 Elf *elf;
 196                 GElf_Ehdr ehdr;
 197                 Elf_Data *symbols;
 198                 size_t strtabidx;
 199                 struct {
 200                         GElf_Shdr shdr;
 201                         Elf_Data *data;
 202                 } *reloc;
 203                 int nr_reloc;
 204         } efile;
 205         /*
 206          * All loaded bpf_object is linked in a list, which is
 207          * hidden to caller. bpf_objects__<func> handlers deal with
 208          * all objects.
 209          */
 210         struct list_head list;
 211         char path[];
 212 };
 213 #define obj_elf_valid(o)        ((o)->efile.elf)
 214
 215 static void bpf_program__unload(struct bpf_program *prog)
 216 {
 217         int i;
 218
 219         if (!prog)
 220                 return;
 221
 222         /*
 223          * If the object is opened but the program was never loaded,
 224          * it is possible that prog->instances.nr == -1.
 225          */
 226         if (prog->instances.nr > 0) {
 227                 for (i = 0; i < prog->instances.nr; i++)
 228                         zclose(prog->instances.fds[i]);
 229         } else if (prog->instances.nr != -1) {
 230                 pr_warning("Internal error: instances.nr is %d\n",
 231                            prog->instances.nr);
 232         }
 233
 234         prog->instances.nr = -1;
 235         zfree(&prog->instances.fds);
 236 }
 237
 238 static void bpf_program__exit(struct bpf_program *prog)
 239 {
 240         if (!prog)
 241                 return;
 242
 243         if (prog->clear_priv)
 244                 prog->clear_priv(prog, prog->priv);
 245
 246         prog->priv = NULL;
 247         prog->clear_priv = NULL;
 248
 249         bpf_program__unload(prog);
 250         zfree(&prog->section_name);
 251         zfree(&prog->insns);
 252         zfree(&prog->reloc_desc);
 253
 254         prog->nr_reloc = 0;
 255         prog->insns_cnt = 0;
 256         prog->idx = -1;
 257 }
 258
 259 static int
 260 bpf_program__init(void *data, size_t size, char *name, int idx,
 261                     struct bpf_program *prog)
 262 {
 263         if (size < sizeof(struct bpf_insn)) {
 264                 pr_warning("corrupted section '%s'\n", name);
 265                 return -EINVAL;
 266         }
 267
 268         bzero(prog, sizeof(*prog));
 269
 270         prog->section_name = strdup(name);
 271         if (!prog->section_name) {
 272                 pr_warning("failed to alloc name for prog %s\n",
 273                            name);
 274                 goto errout;
 275         }
 276
 277         prog->insns = malloc(size);
 278         if (!prog->insns) {
 279                 pr_warning("failed to alloc insns for %s\n", name);
 280                 goto errout;
 281         }
 282         prog->insns_cnt = size / sizeof(struct bpf_insn);
 283         memcpy(prog->insns, data,
 284                prog->insns_cnt * sizeof(struct bpf_insn));
 285         prog->idx = idx;
 286         prog->instances.fds = NULL;
 287         prog->instances.nr = -1;
 288
 289         return 0;
 290 errout:
 291         bpf_program__exit(prog);
 292         return -ENOMEM;
 293 }
 294
 295 static int
 296 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 297                         char *name, int idx)
 298 {
 299         struct bpf_program prog, *progs;
 300         int nr_progs, err;
 301
 302         err = bpf_program__init(data, size, name, idx, &prog);
 303         if (err)
 304                 return err;
 305
 306         progs = obj->programs;
 307         nr_progs = obj->nr_programs;
 308
 309         progs = realloc(progs, sizeof(progs[0]) * (nr_progs + 1));
 310         if (!progs) {
 311                 /*
 312                  * In this case the original obj->programs
 313                  * is still valid, so don't need special treat for
 314                  * bpf_close_object().
 315                  */
 316                 pr_warning("failed to alloc a new program '%s'\n",
 317                            name);
 318                 bpf_program__exit(&prog);
 319                 return -ENOMEM;
 320         }
 321
 322         pr_debug("found program %s\n", prog.section_name);
 323         obj->programs = progs;
 324         obj->nr_programs = nr_progs + 1;
 325         prog.obj = obj;
 326         progs[nr_progs] = prog;
 327         return 0;
 328 }
 329
 330 static struct bpf_object *bpf_object__new(const char *path,
 331                                           void *obj_buf,
 332                                           size_t obj_buf_sz)
 333 {
 334         struct bpf_object *obj;
 335
 336         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 337         if (!obj) {
 338                 pr_warning("alloc memory failed for %s\n", path);
 339                 return ERR_PTR(-ENOMEM);
 340         }
 341
 342         strcpy(obj->path, path);
 343         obj->efile.fd = -1;
 344
 345         /*
 346          * Caller of this function should also calls
 347          * bpf_object__elf_finish() after data collection to return
 348          * obj_buf to user. If not, we should duplicate the buffer to
 349          * avoid user freeing them before elf finish.
 350          */
 351         obj->efile.obj_buf = obj_buf;
 352         obj->efile.obj_buf_sz = obj_buf_sz;
 353
 354         obj->loaded = false;
 355
 356         INIT_LIST_HEAD(&obj->list);
 357         list_add(&obj->list, &bpf_objects_list);
 358         return obj;
 359 }
 360
 361 static void bpf_object__elf_finish(struct bpf_object *obj)
 362 {
 363         if (!obj_elf_valid(obj))
 364                 return;
 365
 366         if (obj->efile.elf) {
 367                 elf_end(obj->efile.elf);
 368                 obj->efile.elf = NULL;
 369         }
 370         obj->efile.symbols = NULL;
 371
 372         zfree(&obj->efile.reloc);
 373         obj->efile.nr_reloc = 0;
 374         zclose(obj->efile.fd);
 375         obj->efile.obj_buf = NULL;
 376         obj->efile.obj_buf_sz = 0;
 377 }
 378
 379 static int bpf_object__elf_init(struct bpf_object *obj)
 380 {
 381         int err = 0;
 382         GElf_Ehdr *ep;
 383
 384         if (obj_elf_valid(obj)) {
 385                 pr_warning("elf init: internal error\n");
 386                 return -LIBBPF_ERRNO__LIBELF;
 387         }
 388
 389         if (obj->efile.obj_buf_sz > 0) {
 390                 /*
 391                  * obj_buf should have been validated by
 392                  * bpf_object__open_buffer().
 393                  */
 394                 obj->efile.elf = elf_memory(obj->efile.obj_buf,
 395                                             obj->efile.obj_buf_sz);
 396         } else {
 397                 obj->efile.fd = open(obj->path, O_RDONLY);
 398                 if (obj->efile.fd < 0) {
 399                         pr_warning("failed to open %s: %s\n", obj->path,
 400                                         strerror(errno));
 401                         return -errno;
 402                 }
 403
 404                 obj->efile.elf = elf_begin(obj->efile.fd,
 405                                 LIBBPF_ELF_C_READ_MMAP,
 406                                 NULL);
 407         }
 408
 409         if (!obj->efile.elf) {
 410                 pr_warning("failed to open %s as ELF file\n",
 411                                 obj->path);
 412                 err = -LIBBPF_ERRNO__LIBELF;
 413                 goto errout;
 414         }
 415
 416         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 417                 pr_warning("failed to get EHDR from %s\n",
 418                                 obj->path);
 419                 err = -LIBBPF_ERRNO__FORMAT;
 420                 goto errout;
 421         }
 422         ep = &obj->efile.ehdr;
 423
 424         if ((ep->e_type != ET_REL) || (ep->e_machine != 0)) {
 425                 pr_warning("%s is not an eBPF object file\n",
 426                         obj->path);
 427                 err = -LIBBPF_ERRNO__FORMAT;
 428                 goto errout;
 429         }
 430
 431         return 0;
 432 errout:
 433         bpf_object__elf_finish(obj);
 434         return err;
 435 }
 436
 437 static int
 438 bpf_object__check_endianness(struct bpf_object *obj)
 439 {
 440         static unsigned int const endian = 1;
 441
 442         switch (obj->efile.ehdr.e_ident[EI_DATA]) {
 443         case ELFDATA2LSB:
 444                 /* We are big endian, BPF obj is little endian. */
 445                 if (*(unsigned char const *)&endian != 1)
 446                         goto mismatch;
 447                 break;
 448
 449         case ELFDATA2MSB:
 450                 /* We are little endian, BPF obj is big endian. */
 451                 if (*(unsigned char const *)&endian != 0)
 452                         goto mismatch;
 453                 break;
 454         default:
 455                 return -LIBBPF_ERRNO__ENDIAN;
 456         }
 457
 458         return 0;
 459
 460 mismatch:
 461         pr_warning("Error: endianness mismatch.\n");
 462         return -LIBBPF_ERRNO__ENDIAN;
 463 }
 464
 465 static int
 466 bpf_object__init_license(struct bpf_object *obj,
 467                          void *data, size_t size)
 468 {
 469         memcpy(obj->license, data,
 470                min(size, sizeof(obj->license) - 1));
 471         pr_debug("license of %s is %s\n", obj->path, obj->license);
 472         return 0;
 473 }
 474
 475 static int
 476 bpf_object__init_kversion(struct bpf_object *obj,
 477                           void *data, size_t size)
 478 {
 479         u32 kver;
 480
 481         if (size != sizeof(kver)) {
 482                 pr_warning("invalid kver section in %s\n", obj->path);
 483                 return -LIBBPF_ERRNO__FORMAT;
 484         }
 485         memcpy(&kver, data, sizeof(kver));
 486         obj->kern_version = kver;
 487         pr_debug("kernel version of %s is %x\n", obj->path,
 488                  obj->kern_version);
 489         return 0;
 490 }
 491
 492 static int
 493 bpf_object__init_maps(struct bpf_object *obj, void *data,
 494                       size_t size)
 495 {
 496         size_t nr_maps;
 497         int i;
 498
 499         nr_maps = size / sizeof(struct bpf_map_def);
 500         if (!data || !nr_maps) {
 501                 pr_debug("%s doesn't need map definition\n",
 502                          obj->path);
 503                 return 0;
 504         }
 505
 506         pr_debug("maps in %s: %zd bytes\n", obj->path, size);
 507
 508         obj->maps = calloc(nr_maps, sizeof(obj->maps[0]));
 509         if (!obj->maps) {
 510                 pr_warning("alloc maps for object failed\n");
 511                 return -ENOMEM;
 512         }
 513         obj->nr_maps = nr_maps;
 514
 515         for (i = 0; i < nr_maps; i++) {
 516                 struct bpf_map_def *def = &obj->maps[i].def;
 517
 518                 /*
 519                  * fill all fd with -1 so won't close incorrect
 520                  * fd (fd=0 is stdin) when failure (zclose won't close
 521                  * negative fd)).
 522                  */
 523                 obj->maps[i].fd = -1;
 524
 525                 /* Save map definition into obj->maps */
 526                 *def = ((struct bpf_map_def *)data)[i];
 527         }
 528         return 0;
 529 }
 530
 531 static int
 532 bpf_object__init_maps_name(struct bpf_object *obj, int maps_shndx)
 533 {
 534         int i;
 535         Elf_Data *symbols = obj->efile.symbols;
 536
 537         if (!symbols || maps_shndx < 0)
 538                 return -EINVAL;
 539
 540         for (i = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 541                 GElf_Sym sym;
 542                 size_t map_idx;
 543                 const char *map_name;
 544
 545                 if (!gelf_getsym(symbols, i, &sym))
 546                         continue;
 547                 if (sym.st_shndx != maps_shndx)
 548                         continue;
 549
 550                 map_name = elf_strptr(obj->efile.elf,
 551                                       obj->efile.strtabidx,
 552                                       sym.st_name);
 553                 map_idx = sym.st_value / sizeof(struct bpf_map_def);
 554                 if (map_idx >= obj->nr_maps) {
 555                         pr_warning("index of map \"%s\" is buggy: %zu > %zu\n",
 556                                    map_name, map_idx, obj->nr_maps);
 557                         continue;
 558                 }
 559                 obj->maps[map_idx].name = strdup(map_name);
 560                 if (!obj->maps[map_idx].name) {
 561                         pr_warning("failed to alloc map name\n");
 562                         return -ENOMEM;
 563                 }
 564                 pr_debug("map %zu is \"%s\"\n", map_idx,
 565                          obj->maps[map_idx].name);
 566         }
 567         return 0;
 568 }
 569
 570 static int bpf_object__elf_collect(struct bpf_object *obj)
 571 {
 572         Elf *elf = obj->efile.elf;
 573         GElf_Ehdr *ep = &obj->efile.ehdr;
 574         Elf_Scn *scn = NULL;
 575         int idx = 0, err = 0, maps_shndx = -1;
 576
 577         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
 578         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
 579                 pr_warning("failed to get e_shstrndx from %s\n",
 580                            obj->path);
 581                 return -LIBBPF_ERRNO__FORMAT;
 582         }
 583
 584         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 585                 char *name;
 586                 GElf_Shdr sh;
 587                 Elf_Data *data;
 588
 589                 idx++;
 590                 if (gelf_getshdr(scn, &sh) != &sh) {
 591                         pr_warning("failed to get section header from %s\n",
 592                                    obj->path);
 593                         err = -LIBBPF_ERRNO__FORMAT;
 594                         goto out;
 595                 }
 596
 597                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 598                 if (!name) {
 599                         pr_warning("failed to get section name from %s\n",
 600                                    obj->path);
 601                         err = -LIBBPF_ERRNO__FORMAT;
 602                         goto out;
 603                 }
 604
 605                 data = elf_getdata(scn, 0);
 606                 if (!data) {
 607                         pr_warning("failed to get section data from %s(%s)\n",
 608                                    name, obj->path);
 609                         err = -LIBBPF_ERRNO__FORMAT;
 610                         goto out;
 611                 }
 612                 pr_debug("section %s, size %ld, link %d, flags %lx, type=%d\n",
 613                          name, (unsigned long)data->d_size,
 614                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
 615                          (int)sh.sh_type);
 616
 617                 if (strcmp(name, "license") == 0)
 618                         err = bpf_object__init_license(obj,
 619                                                        data->d_buf,
 620                                                        data->d_size);
 621                 else if (strcmp(name, "version") == 0)
 622                         err = bpf_object__init_kversion(obj,
 623                                                         data->d_buf,
 624                                                         data->d_size);
 625                 else if (strcmp(name, "maps") == 0) {
 626                         err = bpf_object__init_maps(obj, data->d_buf,
 627                                                     data->d_size);
 628                         maps_shndx = idx;
 629                 } else if (sh.sh_type == SHT_SYMTAB) {
 630                         if (obj->efile.symbols) {
 631                                 pr_warning("bpf: multiple SYMTAB in %s\n",
 632                                            obj->path);
 633                                 err = -LIBBPF_ERRNO__FORMAT;
 634                         } else {
 635                                 obj->efile.symbols = data;
 636                                 obj->efile.strtabidx = sh.sh_link;
 637                         }
 638                 } else if ((sh.sh_type == SHT_PROGBITS) &&
 639                            (sh.sh_flags & SHF_EXECINSTR) &&
 640                            (data->d_size > 0)) {
 641                         err = bpf_object__add_program(obj, data->d_buf,
 642                                                       data->d_size, name, idx);
 643                         if (err) {
 644                                 char errmsg[STRERR_BUFSIZE];
 645
 646                                 strerror_r(-err, errmsg, sizeof(errmsg));
 647                                 pr_warning("failed to alloc program %s (%s): %s",
 648                                            name, obj->path, errmsg);
 649                         }
 650                 } else if (sh.sh_type == SHT_REL) {
 651                         void *reloc = obj->efile.reloc;
 652                         int nr_reloc = obj->efile.nr_reloc + 1;
 653
 654                         reloc = realloc(reloc,
 655                                         sizeof(*obj->efile.reloc) * nr_reloc);
 656                         if (!reloc) {
 657                                 pr_warning("realloc failed\n");
 658                                 err = -ENOMEM;
 659                         } else {
 660                                 int n = nr_reloc - 1;
 661
 662                                 obj->efile.reloc = reloc;
 663                                 obj->efile.nr_reloc = nr_reloc;
 664
 665                                 obj->efile.reloc[n].shdr = sh;
 666                                 obj->efile.reloc[n].data = data;
 667                         }
 668                 }
 669                 if (err)
 670                         goto out;
 671         }
 672
 673         if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
 674                 pr_warning("Corrupted ELF file: index of strtab invalid\n");
 675                 return LIBBPF_ERRNO__FORMAT;
 676         }
 677         if (maps_shndx >= 0)
 678                 err = bpf_object__init_maps_name(obj, maps_shndx);
 679 out:
 680         return err;
 681 }
 682
 683 static struct bpf_program *
 684 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
 685 {
 686         struct bpf_program *prog;
 687         size_t i;
 688
 689         for (i = 0; i < obj->nr_programs; i++) {
 690                 prog = &obj->programs[i];
 691                 if (prog->idx == idx)
 692                         return prog;
 693         }
 694         return NULL;
 695 }
 696
 697 static int
 698 bpf_program__collect_reloc(struct bpf_program *prog,
 699                            size_t nr_maps, GElf_Shdr *shdr,
 700                            Elf_Data *data, Elf_Data *symbols)
 701 {
 702         int i, nrels;
 703
 704         pr_debug("collecting relocating info for: '%s'\n",
 705                  prog->section_name);
 706         nrels = shdr->sh_size / shdr->sh_entsize;
 707
 708         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
 709         if (!prog->reloc_desc) {
 710                 pr_warning("failed to alloc memory in relocation\n");
 711                 return -ENOMEM;
 712         }
 713         prog->nr_reloc = nrels;
 714
 715         for (i = 0; i < nrels; i++) {
 716                 GElf_Sym sym;
 717                 GElf_Rel rel;
 718                 unsigned int insn_idx;
 719                 struct bpf_insn *insns = prog->insns;
 720                 size_t map_idx;
 721
 722                 if (!gelf_getrel(data, i, &rel)) {
 723                         pr_warning("relocation: failed to get %d reloc\n", i);
 724                         return -LIBBPF_ERRNO__FORMAT;
 725                 }
 726
 727                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
 728                 pr_debug("relocation: insn_idx=%u\n", insn_idx);
 729
 730                 if (!gelf_getsym(symbols,
 731                                  GELF_R_SYM(rel.r_info),
 732                                  &sym)) {
 733                         pr_warning("relocation: symbol %"PRIx64" not found\n",
 734                                    GELF_R_SYM(rel.r_info));
 735                         return -LIBBPF_ERRNO__FORMAT;
 736                 }
 737
 738                 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
 739                         pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
 740                                    insn_idx, insns[insn_idx].code);
 741                         return -LIBBPF_ERRNO__RELOC;
 742                 }
 743
 744                 map_idx = sym.st_value / sizeof(struct bpf_map_def);
 745                 if (map_idx >= nr_maps) {
 746                         pr_warning("bpf relocation: map_idx %d large than %d\n",
 747                                    (int)map_idx, (int)nr_maps - 1);
 748                         return -LIBBPF_ERRNO__RELOC;
 749                 }
 750
 751                 prog->reloc_desc[i].insn_idx = insn_idx;
 752                 prog->reloc_desc[i].map_idx = map_idx;
 753         }
 754         return 0;
 755 }
 756
 757 static int
 758 bpf_object__create_maps(struct bpf_object *obj)
 759 {
 760         unsigned int i;
 761
 762         for (i = 0; i < obj->nr_maps; i++) {
 763                 struct bpf_map_def *def = &obj->maps[i].def;
 764                 int *pfd = &obj->maps[i].fd;
 765
 766                 *pfd = bpf_create_map(def->type,
 767                                       def->key_size,
 768                                       def->value_size,
 769                                       def->max_entries);
 770                 if (*pfd < 0) {
 771                         size_t j;
 772                         int err = *pfd;
 773
 774                         pr_warning("failed to create map: %s\n",
 775                                    strerror(errno));
 776                         for (j = 0; j < i; j++)
 777                                 zclose(obj->maps[j].fd);
 778                         return err;
 779                 }
 780                 pr_debug("create map: fd=%d\n", *pfd);
 781         }
 782
 783         return 0;
 784 }
 785
 786 static int
 787 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
 788 {
 789         int i;
 790
 791         if (!prog || !prog->reloc_desc)
 792                 return 0;
 793
 794         for (i = 0; i < prog->nr_reloc; i++) {
 795                 int insn_idx, map_idx;
 796                 struct bpf_insn *insns = prog->insns;
 797
 798                 insn_idx = prog->reloc_desc[i].insn_idx;
 799                 map_idx = prog->reloc_desc[i].map_idx;
 800
 801                 if (insn_idx >= (int)prog->insns_cnt) {
 802                         pr_warning("relocation out of range: '%s'\n",
 803                                    prog->section_name);
 804                         return -LIBBPF_ERRNO__RELOC;
 805                 }
 806                 insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
 807                 insns[insn_idx].imm = obj->maps[map_idx].fd;
 808         }
 809
 810         zfree(&prog->reloc_desc);
 811         prog->nr_reloc = 0;
 812         return 0;
 813 }
 814
 815
 816 static int
 817 bpf_object__relocate(struct bpf_object *obj)
 818 {
 819         struct bpf_program *prog;
 820         size_t i;
 821         int err;
 822
 823         for (i = 0; i < obj->nr_programs; i++) {
 824                 prog = &obj->programs[i];
 825
 826                 err = bpf_program__relocate(prog, obj);
 827                 if (err) {
 828                         pr_warning("failed to relocate '%s'\n",
 829                                    prog->section_name);
 830                         return err;
 831                 }
 832         }
 833         return 0;
 834 }
 835
 836 static int bpf_object__collect_reloc(struct bpf_object *obj)
 837 {
 838         int i, err;
 839
 840         if (!obj_elf_valid(obj)) {
 841                 pr_warning("Internal error: elf object is closed\n");
 842                 return -LIBBPF_ERRNO__INTERNAL;
 843         }
 844
 845         for (i = 0; i < obj->efile.nr_reloc; i++) {
 846                 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
 847                 Elf_Data *data = obj->efile.reloc[i].data;
 848                 int idx = shdr->sh_info;
 849                 struct bpf_program *prog;
 850                 size_t nr_maps = obj->nr_maps;
 851
 852                 if (shdr->sh_type != SHT_REL) {
 853                         pr_warning("internal error at %d\n", __LINE__);
 854                         return -LIBBPF_ERRNO__INTERNAL;
 855                 }
 856
 857                 prog = bpf_object__find_prog_by_idx(obj, idx);
 858                 if (!prog) {
 859                         pr_warning("relocation failed: no %d section\n",
 860                                    idx);
 861                         return -LIBBPF_ERRNO__RELOC;
 862                 }
 863
 864                 err = bpf_program__collect_reloc(prog, nr_maps,
 865                                                  shdr, data,
 866                                                  obj->efile.symbols);
 867                 if (err)
 868                         return err;
 869         }
 870         return 0;
 871 }
 872
 873 static int
 874 load_program(struct bpf_insn *insns, int insns_cnt,
 875              char *license, u32 kern_version, int *pfd)
 876 {
 877         int ret;
 878         char *log_buf;
 879
 880         if (!insns || !insns_cnt)
 881                 return -EINVAL;
 882
 883         log_buf = malloc(BPF_LOG_BUF_SIZE);
 884         if (!log_buf)
 885                 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
 886
 887         ret = bpf_load_program(BPF_PROG_TYPE_KPROBE, insns,
 888                                insns_cnt, license, kern_version,
 889                                log_buf, BPF_LOG_BUF_SIZE);
 890
 891         if (ret >= 0) {
 892                 *pfd = ret;
 893                 ret = 0;
 894                 goto out;
 895         }
 896
 897         ret = -LIBBPF_ERRNO__LOAD;
 898         pr_warning("load bpf program failed: %s\n", strerror(errno));
 899
 900         if (log_buf && log_buf[0] != '\0') {
 901                 ret = -LIBBPF_ERRNO__VERIFY;
 902                 pr_warning("-- BEGIN DUMP LOG ---\n");
 903                 pr_warning("\n%s\n", log_buf);
 904                 pr_warning("-- END LOG --\n");
 905         } else {
 906                 if (insns_cnt >= BPF_MAXINSNS) {
 907                         pr_warning("Program too large (%d insns), at most %d insns\n",
 908                                    insns_cnt, BPF_MAXINSNS);
 909                         ret = -LIBBPF_ERRNO__PROG2BIG;
 910                 } else if (log_buf) {
 911                         pr_warning("log buffer is empty\n");
 912                         ret = -LIBBPF_ERRNO__KVER;
 913                 }
 914         }
 915
 916 out:
 917         free(log_buf);
 918         return ret;
 919 }
 920
 921 static int
 922 bpf_program__load(struct bpf_program *prog,
 923                   char *license, u32 kern_version)
 924 {
 925         int err = 0, fd, i;
 926
 927         if (prog->instances.nr < 0 || !prog->instances.fds) {
 928                 if (prog->preprocessor) {
 929                         pr_warning("Internal error: can't load program '%s'\n",
 930                                    prog->section_name);
 931                         return -LIBBPF_ERRNO__INTERNAL;
 932                 }
 933
 934                 prog->instances.fds = malloc(sizeof(int));
 935                 if (!prog->instances.fds) {
 936                         pr_warning("Not enough memory for BPF fds\n");
 937                         return -ENOMEM;
 938                 }
 939                 prog->instances.nr = 1;
 940                 prog->instances.fds[0] = -1;
 941         }
 942
 943         if (!prog->preprocessor) {
 944                 if (prog->instances.nr != 1) {
 945                         pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
 946                                    prog->section_name, prog->instances.nr);
 947                 }
 948                 err = load_program(prog->insns, prog->insns_cnt,
 949                                    license, kern_version, &fd);
 950                 if (!err)
 951                         prog->instances.fds[0] = fd;
 952                 goto out;
 953         }
 954
 955         for (i = 0; i < prog->instances.nr; i++) {
 956                 struct bpf_prog_prep_result result;
 957                 bpf_program_prep_t preprocessor = prog->preprocessor;
 958
 959                 bzero(&result, sizeof(result));
 960                 err = preprocessor(prog, i, prog->insns,
 961                                    prog->insns_cnt, &result);
 962                 if (err) {
 963                         pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
 964                                    i, prog->section_name);
 965                         goto out;
 966                 }
 967
 968                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
 969                         pr_debug("Skip loading the %dth instance of program '%s'\n",
 970                                  i, prog->section_name);
 971                         prog->instances.fds[i] = -1;
 972                         if (result.pfd)
 973                                 *result.pfd = -1;
 974                         continue;
 975                 }
 976
 977                 err = load_program(result.new_insn_ptr,
 978                                    result.new_insn_cnt,
 979                                    license, kern_version, &fd);
 980
 981                 if (err) {
 982                         pr_warning("Loading the %dth instance of program '%s' failed\n",
 983                                         i, prog->section_name);
 984                         goto out;
 985                 }
 986
 987                 if (result.pfd)
 988                         *result.pfd = fd;
 989                 prog->instances.fds[i] = fd;
 990         }
 991 out:
 992         if (err)
 993                 pr_warning("failed to load program '%s'\n",
 994                            prog->section_name);
 995         zfree(&prog->insns);
 996         prog->insns_cnt = 0;
 997         return err;
 998 }
 999
1000 static int
1001 bpf_object__load_progs(struct bpf_object *obj)
1002 {
1003         size_t i;
1004         int err;
1005
1006         for (i = 0; i < obj->nr_programs; i++) {
1007                 err = bpf_program__load(&obj->programs[i],
1008                                         obj->license,
1009                                         obj->kern_version);
1010                 if (err)
1011                         return err;
1012         }
1013         return 0;
1014 }
1015
1016 static int bpf_object__validate(struct bpf_object *obj)
1017 {
1018         if (obj->kern_version == 0) {
1019                 pr_warning("%s doesn't provide kernel version\n",
1020                            obj->path);
1021                 return -LIBBPF_ERRNO__KVERSION;
1022         }
1023         return 0;
1024 }
1025
1026 static struct bpf_object *
1027 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz)
1028 {
1029         struct bpf_object *obj;
1030         int err;
1031
1032         if (elf_version(EV_CURRENT) == EV_NONE) {
1033                 pr_warning("failed to init libelf for %s\n", path);
1034                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1035         }
1036
1037         obj = bpf_object__new(path, obj_buf, obj_buf_sz);
1038         if (IS_ERR(obj))
1039                 return obj;
1040
1041         CHECK_ERR(bpf_object__elf_init(obj), err, out);
1042         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
1043         CHECK_ERR(bpf_object__elf_collect(obj), err, out);
1044         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
1045         CHECK_ERR(bpf_object__validate(obj), err, out);
1046
1047         bpf_object__elf_finish(obj);
1048         return obj;
1049 out:
1050         bpf_object__close(obj);
1051         return ERR_PTR(err);
1052 }
1053
1054 struct bpf_object *bpf_object__open(const char *path)
1055 {
1056         /* param validation */
1057         if (!path)
1058                 return NULL;
1059
1060         pr_debug("loading %s\n", path);
1061
1062         return __bpf_object__open(path, NULL, 0);
1063 }
1064
1065 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
1066                                            size_t obj_buf_sz,
1067                                            const char *name)
1068 {
1069         char tmp_name[64];
1070
1071         /* param validation */
1072         if (!obj_buf || obj_buf_sz <= 0)
1073                 return NULL;
1074
1075         if (!name) {
1076                 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
1077                          (unsigned long)obj_buf,
1078                          (unsigned long)obj_buf_sz);
1079                 tmp_name[sizeof(tmp_name) - 1] = '\0';
1080                 name = tmp_name;
1081         }
1082         pr_debug("loading object '%s' from buffer\n",
1083                  name);
1084
1085         return __bpf_object__open(name, obj_buf, obj_buf_sz);
1086 }
1087
1088 int bpf_object__unload(struct bpf_object *obj)
1089 {
1090         size_t i;
1091
1092         if (!obj)
1093                 return -EINVAL;
1094
1095         for (i = 0; i < obj->nr_maps; i++)
1096                 zclose(obj->maps[i].fd);
1097
1098         for (i = 0; i < obj->nr_programs; i++)
1099                 bpf_program__unload(&obj->programs[i]);
1100
1101         return 0;
1102 }
1103
1104 int bpf_object__load(struct bpf_object *obj)
1105 {
1106         int err;
1107
1108         if (!obj)
1109                 return -EINVAL;
1110
1111         if (obj->loaded) {
1112                 pr_warning("object should not be loaded twice\n");
1113                 return -EINVAL;
1114         }
1115
1116         obj->loaded = true;
1117
1118         CHECK_ERR(bpf_object__create_maps(obj), err, out);
1119         CHECK_ERR(bpf_object__relocate(obj), err, out);
1120         CHECK_ERR(bpf_object__load_progs(obj), err, out);
1121
1122         return 0;
1123 out:
1124         bpf_object__unload(obj);
1125         pr_warning("failed to load object '%s'\n", obj->path);
1126         return err;
1127 }
1128
1129 void bpf_object__close(struct bpf_object *obj)
1130 {
1131         size_t i;
1132
1133         if (!obj)
1134                 return;
1135
1136         bpf_object__elf_finish(obj);
1137         bpf_object__unload(obj);
1138
1139         for (i = 0; i < obj->nr_maps; i++) {
1140                 zfree(&obj->maps[i].name);
1141                 if (obj->maps[i].clear_priv)
1142                         obj->maps[i].clear_priv(&obj->maps[i],
1143                                                 obj->maps[i].priv);
1144                 obj->maps[i].priv = NULL;
1145                 obj->maps[i].clear_priv = NULL;
1146         }
1147         zfree(&obj->maps);
1148         obj->nr_maps = 0;
1149
1150         if (obj->programs && obj->nr_programs) {
1151                 for (i = 0; i < obj->nr_programs; i++)
1152                         bpf_program__exit(&obj->programs[i]);
1153         }
1154         zfree(&obj->programs);
1155
1156         list_del(&obj->list);
1157         free(obj);
1158 }
1159
1160 struct bpf_object *
1161 bpf_object__next(struct bpf_object *prev)
1162 {
1163         struct bpf_object *next;
1164
1165         if (!prev)
1166                 next = list_first_entry(&bpf_objects_list,
1167                                         struct bpf_object,
1168                                         list);
1169         else
1170                 next = list_next_entry(prev, list);
1171
1172         /* Empty list is noticed here so don't need checking on entry. */
1173         if (&next->list == &bpf_objects_list)
1174                 return NULL;
1175
1176         return next;
1177 }
1178
1179 const char *
1180 bpf_object__get_name(struct bpf_object *obj)
1181 {
1182         if (!obj)
1183                 return ERR_PTR(-EINVAL);
1184         return obj->path;
1185 }
1186
1187 unsigned int
1188 bpf_object__get_kversion(struct bpf_object *obj)
1189 {
1190         if (!obj)
1191                 return 0;
1192         return obj->kern_version;
1193 }
1194
1195 struct bpf_program *
1196 bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1197 {
1198         size_t idx;
1199
1200         if (!obj->programs)
1201                 return NULL;
1202         /* First handler */
1203         if (prev == NULL)
1204                 return &obj->programs[0];
1205
1206         if (prev->obj != obj) {
1207                 pr_warning("error: program handler doesn't match object\n");
1208                 return NULL;
1209         }
1210
1211         idx = (prev - obj->programs) + 1;
1212         if (idx >= obj->nr_programs)
1213                 return NULL;
1214         return &obj->programs[idx];
1215 }
1216
1217 int bpf_program__set_private(struct bpf_program *prog,
1218                              void *priv,
1219                              bpf_program_clear_priv_t clear_priv)
1220 {
1221         if (prog->priv && prog->clear_priv)
1222                 prog->clear_priv(prog, prog->priv);
1223
1224         prog->priv = priv;
1225         prog->clear_priv = clear_priv;
1226         return 0;
1227 }
1228
1229 int bpf_program__get_private(struct bpf_program *prog, void **ppriv)
1230 {
1231         *ppriv = prog->priv;
1232         return 0;
1233 }
1234
1235 const char *bpf_program__title(struct bpf_program *prog, bool needs_copy)
1236 {
1237         const char *title;
1238
1239         title = prog->section_name;
1240         if (needs_copy) {
1241                 title = strdup(title);
1242                 if (!title) {
1243                         pr_warning("failed to strdup program title\n");
1244                         return ERR_PTR(-ENOMEM);
1245                 }
1246         }
1247
1248         return title;
1249 }
1250
1251 int bpf_program__fd(struct bpf_program *prog)
1252 {
1253         return bpf_program__nth_fd(prog, 0);
1254 }
1255
1256 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
1257                           bpf_program_prep_t prep)
1258 {
1259         int *instances_fds;
1260
1261         if (nr_instances <= 0 || !prep)
1262                 return -EINVAL;
1263
1264         if (prog->instances.nr > 0 || prog->instances.fds) {
1265                 pr_warning("Can't set pre-processor after loading\n");
1266                 return -EINVAL;
1267         }
1268
1269         instances_fds = malloc(sizeof(int) * nr_instances);
1270         if (!instances_fds) {
1271                 pr_warning("alloc memory failed for fds\n");
1272                 return -ENOMEM;
1273         }
1274
1275         /* fill all fd with -1 */
1276         memset(instances_fds, -1, sizeof(int) * nr_instances);
1277
1278         prog->instances.nr = nr_instances;
1279         prog->instances.fds = instances_fds;
1280         prog->preprocessor = prep;
1281         return 0;
1282 }
1283
1284 int bpf_program__nth_fd(struct bpf_program *prog, int n)
1285 {
1286         int fd;
1287
1288         if (n >= prog->instances.nr || n < 0) {
1289                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
1290                            n, prog->section_name, prog->instances.nr);
1291                 return -EINVAL;
1292         }
1293
1294         fd = prog->instances.fds[n];
1295         if (fd < 0) {
1296                 pr_warning("%dth instance of program '%s' is invalid\n",
1297                            n, prog->section_name);
1298                 return -ENOENT;
1299         }
1300
1301         return fd;
1302 }
1303
1304 int bpf_map__get_fd(struct bpf_map *map)
1305 {
1306         if (!map)
1307                 return -EINVAL;
1308
1309         return map->fd;
1310 }
1311
1312 int bpf_map__get_def(struct bpf_map *map, struct bpf_map_def *pdef)
1313 {
1314         if (!map || !pdef)
1315                 return -EINVAL;
1316
1317         *pdef = map->def;
1318         return 0;
1319 }
1320
1321 const char *bpf_map__get_name(struct bpf_map *map)
1322 {
1323         if (!map)
1324                 return NULL;
1325         return map->name;
1326 }
1327
1328 int bpf_map__set_private(struct bpf_map *map, void *priv,
1329                          bpf_map_clear_priv_t clear_priv)
1330 {
1331         if (!map)
1332                 return -EINVAL;
1333
1334         if (map->priv) {
1335                 if (map->clear_priv)
1336                         map->clear_priv(map, map->priv);
1337         }
1338
1339         map->priv = priv;
1340         map->clear_priv = clear_priv;
1341         return 0;
1342 }
1343
1344 int bpf_map__get_private(struct bpf_map *map, void **ppriv)
1345 {
1346         if (!map)
1347                 return -EINVAL;
1348
1349         if (ppriv)
1350                 *ppriv = map->priv;
1351         return 0;
1352 }
1353
1354 struct bpf_map *
1355 bpf_map__next(struct bpf_map *prev, struct bpf_object *obj)
1356 {
1357         size_t idx;
1358         struct bpf_map *s, *e;
1359
1360         if (!obj || !obj->maps)
1361                 return NULL;
1362
1363         s = obj->maps;
1364         e = obj->maps + obj->nr_maps;
1365
1366         if (prev == NULL)
1367                 return s;
1368
1369         if ((prev < s) || (prev >= e)) {
1370                 pr_warning("error in %s: map handler doesn't belong to object\n",
1371                            __func__);
1372                 return NULL;
1373         }
1374
1375         idx = (prev - obj->maps) + 1;
1376         if (idx >= obj->nr_maps)
1377                 return NULL;
1378         return &obj->maps[idx];
1379 }
1380
1381 struct bpf_map *
1382 bpf_object__get_map_by_name(struct bpf_object *obj, const char *name)
1383 {
1384         struct bpf_map *pos;
1385
1386         bpf_map__for_each(pos, obj) {
1387                 if (pos->name && !strcmp(pos->name, name))
1388                         return pos;
1389         }
1390         return NULL;
1391 }