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