tools/lib/bpf/libbpf.c

   1 // SPDX-License-Identifier: LGPL-2.1
   2
   3 /*
   4  * Common eBPF ELF object loading operations.
   5  *
   6  * Copyright (C) 2013-2015 Alexei Starovoitov <ast@kernel.org>
   7  * Copyright (C) 2015 Wang Nan <wangnan0@huawei.com>
   8  * Copyright (C) 2015 Huawei Inc.
   9  * Copyright (C) 2017 Nicira, Inc.
  10  *
  11  * This program is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public
  13  * License as published by the Free Software Foundation;
  14  * version 2.1 of the License (not later!)
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19  * GNU Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with this program; if not,  see <http://www.gnu.org/licenses>
  23  */
  24
  25 #include <stdlib.h>
  26 #include <stdio.h>
  27 #include <stdarg.h>
  28 #include <libgen.h>
  29 #include <inttypes.h>
  30 #include <string.h>
  31 #include <unistd.h>
  32 #include <fcntl.h>
  33 #include <errno.h>
  34 #include <asm/unistd.h>
  35 #include <linux/err.h>
  36 #include <linux/kernel.h>
  37 #include <linux/bpf.h>
  38 #include <linux/list.h>
  39 #include <linux/limits.h>
  40 #include <sys/stat.h>
  41 #include <sys/types.h>
  42 #include <sys/vfs.h>
  43 #include <libelf.h>
  44 #include <gelf.h>
  45
  46 #include "libbpf.h"
  47 #include "bpf.h"
  48
  49 #ifndef EM_BPF
  50 #define EM_BPF 247
  51 #endif
  52
  53 #ifndef BPF_FS_MAGIC
  54 #define BPF_FS_MAGIC            0xcafe4a11
  55 #endif
  56
  57 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  58
  59 __printf(1, 2)
  60 static int __base_pr(const char *format, ...)
  61 {
  62         va_list args;
  63         int err;
  64
  65         va_start(args, format);
  66         err = vfprintf(stderr, format, args);
  67         va_end(args);
  68         return err;
  69 }
  70
  71 static __printf(1, 2) libbpf_print_fn_t __pr_warning = __base_pr;
  72 static __printf(1, 2) libbpf_print_fn_t __pr_info = __base_pr;
  73 static __printf(1, 2) libbpf_print_fn_t __pr_debug;
  74
  75 #define __pr(func, fmt, ...)    \
  76 do {                            \
  77         if ((func))             \
  78                 (func)("libbpf: " fmt, ##__VA_ARGS__); \
  79 } while (0)
  80
  81 #define pr_warning(fmt, ...)    __pr(__pr_warning, fmt, ##__VA_ARGS__)
  82 #define pr_info(fmt, ...)       __pr(__pr_info, fmt, ##__VA_ARGS__)
  83 #define pr_debug(fmt, ...)      __pr(__pr_debug, fmt, ##__VA_ARGS__)
  84
  85 void libbpf_set_print(libbpf_print_fn_t warn,
  86                       libbpf_print_fn_t info,
  87                       libbpf_print_fn_t debug)
  88 {
  89         __pr_warning = warn;
  90         __pr_info = info;
  91         __pr_debug = debug;
  92 }
  93
  94 #define STRERR_BUFSIZE  128
  95
  96 #define ERRNO_OFFSET(e)         ((e) - __LIBBPF_ERRNO__START)
  97 #define ERRCODE_OFFSET(c)       ERRNO_OFFSET(LIBBPF_ERRNO__##c)
  98 #define NR_ERRNO        (__LIBBPF_ERRNO__END - __LIBBPF_ERRNO__START)
  99
 100 static const char *libbpf_strerror_table[NR_ERRNO] = {
 101         [ERRCODE_OFFSET(LIBELF)]        = "Something wrong in libelf",
 102         [ERRCODE_OFFSET(FORMAT)]        = "BPF object format invalid",
 103         [ERRCODE_OFFSET(KVERSION)]      = "'version' section incorrect or lost",
 104         [ERRCODE_OFFSET(ENDIAN)]        = "Endian mismatch",
 105         [ERRCODE_OFFSET(INTERNAL)]      = "Internal error in libbpf",
 106         [ERRCODE_OFFSET(RELOC)]         = "Relocation failed",
 107         [ERRCODE_OFFSET(VERIFY)]        = "Kernel verifier blocks program loading",
 108         [ERRCODE_OFFSET(PROG2BIG)]      = "Program too big",
 109         [ERRCODE_OFFSET(KVER)]          = "Incorrect kernel version",
 110         [ERRCODE_OFFSET(PROGTYPE)]      = "Kernel doesn't support this program type",
 111         [ERRCODE_OFFSET(WRNGPID)]       = "Wrong pid in netlink message",
 112         [ERRCODE_OFFSET(INVSEQ)]        = "Invalid netlink sequence",
 113 };
 114
 115 int libbpf_strerror(int err, char *buf, size_t size)
 116 {
 117         if (!buf || !size)
 118                 return -1;
 119
 120         err = err > 0 ? err : -err;
 121
 122         if (err < __LIBBPF_ERRNO__START) {
 123                 int ret;
 124
 125                 ret = strerror_r(err, buf, size);
 126                 buf[size - 1] = '\0';
 127                 return ret;
 128         }
 129
 130         if (err < __LIBBPF_ERRNO__END) {
 131                 const char *msg;
 132
 133                 msg = libbpf_strerror_table[ERRNO_OFFSET(err)];
 134                 snprintf(buf, size, "%s", msg);
 135                 buf[size - 1] = '\0';
 136                 return 0;
 137         }
 138
 139         snprintf(buf, size, "Unknown libbpf error %d", err);
 140         buf[size - 1] = '\0';
 141         return -1;
 142 }
 143
 144 #define CHECK_ERR(action, err, out) do {        \
 145         err = action;                   \
 146         if (err)                        \
 147                 goto out;               \
 148 } while(0)
 149
 150
 151 /* Copied from tools/perf/util/util.h */
 152 #ifndef zfree
 153 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 154 #endif
 155
 156 #ifndef zclose
 157 # define zclose(fd) ({                  \
 158         int ___err = 0;                 \
 159         if ((fd) >= 0)                  \
 160                 ___err = close((fd));   \
 161         fd = -1;                        \
 162         ___err; })
 163 #endif
 164
 165 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 166 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 167 #else
 168 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 169 #endif
 170
 171 /*
 172  * bpf_prog should be a better name but it has been used in
 173  * linux/filter.h.
 174  */
 175 struct bpf_program {
 176         /* Index in elf obj file, for relocation use. */
 177         int idx;
 178         char *name;
 179         char *section_name;
 180         struct bpf_insn *insns;
 181         size_t insns_cnt, main_prog_cnt;
 182         enum bpf_prog_type type;
 183
 184         struct reloc_desc {
 185                 enum {
 186                         RELO_LD64,
 187                         RELO_CALL,
 188                 } type;
 189                 int insn_idx;
 190                 union {
 191                         int map_idx;
 192                         int text_off;
 193                 };
 194         } *reloc_desc;
 195         int nr_reloc;
 196
 197         struct {
 198                 int nr;
 199                 int *fds;
 200         } instances;
 201         bpf_program_prep_t preprocessor;
 202
 203         struct bpf_object *obj;
 204         void *priv;
 205         bpf_program_clear_priv_t clear_priv;
 206 };
 207
 208 struct bpf_map {
 209         int fd;
 210         char *name;
 211         size_t offset;
 212         struct bpf_map_def def;
 213         void *priv;
 214         bpf_map_clear_priv_t clear_priv;
 215 };
 216
 217 static LIST_HEAD(bpf_objects_list);
 218
 219 struct bpf_object {
 220         char license[64];
 221         u32 kern_version;
 222
 223         struct bpf_program *programs;
 224         size_t nr_programs;
 225         struct bpf_map *maps;
 226         size_t nr_maps;
 227
 228         bool loaded;
 229
 230         /*
 231          * Information when doing elf related work. Only valid if fd
 232          * is valid.
 233          */
 234         struct {
 235                 int fd;
 236                 void *obj_buf;
 237                 size_t obj_buf_sz;
 238                 Elf *elf;
 239                 GElf_Ehdr ehdr;
 240                 Elf_Data *symbols;
 241                 size_t strtabidx;
 242                 struct {
 243                         GElf_Shdr shdr;
 244                         Elf_Data *data;
 245                 } *reloc;
 246                 int nr_reloc;
 247                 int maps_shndx;
 248                 int text_shndx;
 249         } efile;
 250         /*
 251          * All loaded bpf_object is linked in a list, which is
 252          * hidden to caller. bpf_objects__<func> handlers deal with
 253          * all objects.
 254          */
 255         struct list_head list;
 256
 257         void *priv;
 258         bpf_object_clear_priv_t clear_priv;
 259
 260         char path[];
 261 };
 262 #define obj_elf_valid(o)        ((o)->efile.elf)
 263
 264 static void bpf_program__unload(struct bpf_program *prog)
 265 {
 266         int i;
 267
 268         if (!prog)
 269                 return;
 270
 271         /*
 272          * If the object is opened but the program was never loaded,
 273          * it is possible that prog->instances.nr == -1.
 274          */
 275         if (prog->instances.nr > 0) {
 276                 for (i = 0; i < prog->instances.nr; i++)
 277                         zclose(prog->instances.fds[i]);
 278         } else if (prog->instances.nr != -1) {
 279                 pr_warning("Internal error: instances.nr is %d\n",
 280                            prog->instances.nr);
 281         }
 282
 283         prog->instances.nr = -1;
 284         zfree(&prog->instances.fds);
 285 }
 286
 287 static void bpf_program__exit(struct bpf_program *prog)
 288 {
 289         if (!prog)
 290                 return;
 291
 292         if (prog->clear_priv)
 293                 prog->clear_priv(prog, prog->priv);
 294
 295         prog->priv = NULL;
 296         prog->clear_priv = NULL;
 297
 298         bpf_program__unload(prog);
 299         zfree(&prog->name);
 300         zfree(&prog->section_name);
 301         zfree(&prog->insns);
 302         zfree(&prog->reloc_desc);
 303
 304         prog->nr_reloc = 0;
 305         prog->insns_cnt = 0;
 306         prog->idx = -1;
 307 }
 308
 309 static int
 310 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 311                   struct bpf_program *prog)
 312 {
 313         if (size < sizeof(struct bpf_insn)) {
 314                 pr_warning("corrupted section '%s'\n", section_name);
 315                 return -EINVAL;
 316         }
 317
 318         bzero(prog, sizeof(*prog));
 319
 320         prog->section_name = strdup(section_name);
 321         if (!prog->section_name) {
 322                 pr_warning("failed to alloc name for prog under section(%d) %s\n",
 323                            idx, section_name);
 324                 goto errout;
 325         }
 326
 327         prog->insns = malloc(size);
 328         if (!prog->insns) {
 329                 pr_warning("failed to alloc insns for prog under section %s\n",
 330                            section_name);
 331                 goto errout;
 332         }
 333         prog->insns_cnt = size / sizeof(struct bpf_insn);
 334         memcpy(prog->insns, data,
 335                prog->insns_cnt * sizeof(struct bpf_insn));
 336         prog->idx = idx;
 337         prog->instances.fds = NULL;
 338         prog->instances.nr = -1;
 339         prog->type = BPF_PROG_TYPE_KPROBE;
 340
 341         return 0;
 342 errout:
 343         bpf_program__exit(prog);
 344         return -ENOMEM;
 345 }
 346
 347 static int
 348 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 349                         char *section_name, int idx)
 350 {
 351         struct bpf_program prog, *progs;
 352         int nr_progs, err;
 353
 354         err = bpf_program__init(data, size, section_name, idx, &prog);
 355         if (err)
 356                 return err;
 357
 358         progs = obj->programs;
 359         nr_progs = obj->nr_programs;
 360
 361         progs = realloc(progs, sizeof(progs[0]) * (nr_progs + 1));
 362         if (!progs) {
 363                 /*
 364                  * In this case the original obj->programs
 365                  * is still valid, so don't need special treat for
 366                  * bpf_close_object().
 367                  */
 368                 pr_warning("failed to alloc a new program under section '%s'\n",
 369                            section_name);
 370                 bpf_program__exit(&prog);
 371                 return -ENOMEM;
 372         }
 373
 374         pr_debug("found program %s\n", prog.section_name);
 375         obj->programs = progs;
 376         obj->nr_programs = nr_progs + 1;
 377         prog.obj = obj;
 378         progs[nr_progs] = prog;
 379         return 0;
 380 }
 381
 382 static int
 383 bpf_object__init_prog_names(struct bpf_object *obj)
 384 {
 385         Elf_Data *symbols = obj->efile.symbols;
 386         struct bpf_program *prog;
 387         size_t pi, si;
 388
 389         for (pi = 0; pi < obj->nr_programs; pi++) {
 390                 const char *name = NULL;
 391
 392                 prog = &obj->programs[pi];
 393                 if (prog->idx == obj->efile.text_shndx) {
 394                         name = ".text";
 395                         goto skip_search;
 396                 }
 397
 398                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 399                      si++) {
 400                         GElf_Sym sym;
 401
 402                         if (!gelf_getsym(symbols, si, &sym))
 403                                 continue;
 404                         if (sym.st_shndx != prog->idx)
 405                                 continue;
 406                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 407                                 continue;
 408
 409                         name = elf_strptr(obj->efile.elf,
 410                                           obj->efile.strtabidx,
 411                                           sym.st_name);
 412                         if (!name) {
 413                                 pr_warning("failed to get sym name string for prog %s\n",
 414                                            prog->section_name);
 415                                 return -LIBBPF_ERRNO__LIBELF;
 416                         }
 417                 }
 418
 419                 if (!name) {
 420                         pr_warning("failed to find sym for prog %s\n",
 421                                    prog->section_name);
 422                         return -EINVAL;
 423                 }
 424 skip_search:
 425                 prog->name = strdup(name);
 426                 if (!prog->name) {
 427                         pr_warning("failed to allocate memory for prog sym %s\n",
 428                                    name);
 429                         return -ENOMEM;
 430                 }
 431         }
 432
 433         return 0;
 434 }
 435
 436 static struct bpf_object *bpf_object__new(const char *path,
 437                                           void *obj_buf,
 438                                           size_t obj_buf_sz)
 439 {
 440         struct bpf_object *obj;
 441
 442         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 443         if (!obj) {
 444                 pr_warning("alloc memory failed for %s\n", path);
 445                 return ERR_PTR(-ENOMEM);
 446         }
 447
 448         strcpy(obj->path, path);
 449         obj->efile.fd = -1;
 450
 451         /*
 452          * Caller of this function should also calls
 453          * bpf_object__elf_finish() after data collection to return
 454          * obj_buf to user. If not, we should duplicate the buffer to
 455          * avoid user freeing them before elf finish.
 456          */
 457         obj->efile.obj_buf = obj_buf;
 458         obj->efile.obj_buf_sz = obj_buf_sz;
 459         obj->efile.maps_shndx = -1;
 460
 461         obj->loaded = false;
 462
 463         INIT_LIST_HEAD(&obj->list);
 464         list_add(&obj->list, &bpf_objects_list);
 465         return obj;
 466 }
 467
 468 static void bpf_object__elf_finish(struct bpf_object *obj)
 469 {
 470         if (!obj_elf_valid(obj))
 471                 return;
 472
 473         if (obj->efile.elf) {
 474                 elf_end(obj->efile.elf);
 475                 obj->efile.elf = NULL;
 476         }
 477         obj->efile.symbols = NULL;
 478
 479         zfree(&obj->efile.reloc);
 480         obj->efile.nr_reloc = 0;
 481         zclose(obj->efile.fd);
 482         obj->efile.obj_buf = NULL;
 483         obj->efile.obj_buf_sz = 0;
 484 }
 485
 486 static int bpf_object__elf_init(struct bpf_object *obj)
 487 {
 488         int err = 0;
 489         GElf_Ehdr *ep;
 490
 491         if (obj_elf_valid(obj)) {
 492                 pr_warning("elf init: internal error\n");
 493                 return -LIBBPF_ERRNO__LIBELF;
 494         }
 495
 496         if (obj->efile.obj_buf_sz > 0) {
 497                 /*
 498                  * obj_buf should have been validated by
 499                  * bpf_object__open_buffer().
 500                  */
 501                 obj->efile.elf = elf_memory(obj->efile.obj_buf,
 502                                             obj->efile.obj_buf_sz);
 503         } else {
 504                 obj->efile.fd = open(obj->path, O_RDONLY);
 505                 if (obj->efile.fd < 0) {
 506                         pr_warning("failed to open %s: %s\n", obj->path,
 507                                         strerror(errno));
 508                         return -errno;
 509                 }
 510
 511                 obj->efile.elf = elf_begin(obj->efile.fd,
 512                                 LIBBPF_ELF_C_READ_MMAP,
 513                                 NULL);
 514         }
 515
 516         if (!obj->efile.elf) {
 517                 pr_warning("failed to open %s as ELF file\n",
 518                                 obj->path);
 519                 err = -LIBBPF_ERRNO__LIBELF;
 520                 goto errout;
 521         }
 522
 523         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 524                 pr_warning("failed to get EHDR from %s\n",
 525                                 obj->path);
 526                 err = -LIBBPF_ERRNO__FORMAT;
 527                 goto errout;
 528         }
 529         ep = &obj->efile.ehdr;
 530
 531         /* Old LLVM set e_machine to EM_NONE */
 532         if ((ep->e_type != ET_REL) || (ep->e_machine && (ep->e_machine != EM_BPF))) {
 533                 pr_warning("%s is not an eBPF object file\n",
 534                         obj->path);
 535                 err = -LIBBPF_ERRNO__FORMAT;
 536                 goto errout;
 537         }
 538
 539         return 0;
 540 errout:
 541         bpf_object__elf_finish(obj);
 542         return err;
 543 }
 544
 545 static int
 546 bpf_object__check_endianness(struct bpf_object *obj)
 547 {
 548         static unsigned int const endian = 1;
 549
 550         switch (obj->efile.ehdr.e_ident[EI_DATA]) {
 551         case ELFDATA2LSB:
 552                 /* We are big endian, BPF obj is little endian. */
 553                 if (*(unsigned char const *)&endian != 1)
 554                         goto mismatch;
 555                 break;
 556
 557         case ELFDATA2MSB:
 558                 /* We are little endian, BPF obj is big endian. */
 559                 if (*(unsigned char const *)&endian != 0)
 560                         goto mismatch;
 561                 break;
 562         default:
 563                 return -LIBBPF_ERRNO__ENDIAN;
 564         }
 565
 566         return 0;
 567
 568 mismatch:
 569         pr_warning("Error: endianness mismatch.\n");
 570         return -LIBBPF_ERRNO__ENDIAN;
 571 }
 572
 573 static int
 574 bpf_object__init_license(struct bpf_object *obj,
 575                          void *data, size_t size)
 576 {
 577         memcpy(obj->license, data,
 578                min(size, sizeof(obj->license) - 1));
 579         pr_debug("license of %s is %s\n", obj->path, obj->license);
 580         return 0;
 581 }
 582
 583 static int
 584 bpf_object__init_kversion(struct bpf_object *obj,
 585                           void *data, size_t size)
 586 {
 587         u32 kver;
 588
 589         if (size != sizeof(kver)) {
 590                 pr_warning("invalid kver section in %s\n", obj->path);
 591                 return -LIBBPF_ERRNO__FORMAT;
 592         }
 593         memcpy(&kver, data, sizeof(kver));
 594         obj->kern_version = kver;
 595         pr_debug("kernel version of %s is %x\n", obj->path,
 596                  obj->kern_version);
 597         return 0;
 598 }
 599
 600 static int compare_bpf_map(const void *_a, const void *_b)
 601 {
 602         const struct bpf_map *a = _a;
 603         const struct bpf_map *b = _b;
 604
 605         return a->offset - b->offset;
 606 }
 607
 608 static int
 609 bpf_object__init_maps(struct bpf_object *obj)
 610 {
 611         int i, map_idx, map_def_sz, nr_maps = 0;
 612         Elf_Scn *scn;
 613         Elf_Data *data;
 614         Elf_Data *symbols = obj->efile.symbols;
 615
 616         if (obj->efile.maps_shndx < 0)
 617                 return -EINVAL;
 618         if (!symbols)
 619                 return -EINVAL;
 620
 621         scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
 622         if (scn)
 623                 data = elf_getdata(scn, NULL);
 624         if (!scn || !data) {
 625                 pr_warning("failed to get Elf_Data from map section %d\n",
 626                            obj->efile.maps_shndx);
 627                 return -EINVAL;
 628         }
 629
 630         /*
 631          * Count number of maps. Each map has a name.
 632          * Array of maps is not supported: only the first element is
 633          * considered.
 634          *
 635          * TODO: Detect array of map and report error.
 636          */
 637         for (i = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 638                 GElf_Sym sym;
 639
 640                 if (!gelf_getsym(symbols, i, &sym))
 641                         continue;
 642                 if (sym.st_shndx != obj->efile.maps_shndx)
 643                         continue;
 644                 nr_maps++;
 645         }
 646
 647         /* Alloc obj->maps and fill nr_maps. */
 648         pr_debug("maps in %s: %d maps in %zd bytes\n", obj->path,
 649                  nr_maps, data->d_size);
 650
 651         if (!nr_maps)
 652                 return 0;
 653
 654         /* Assume equally sized map definitions */
 655         map_def_sz = data->d_size / nr_maps;
 656         if (!data->d_size || (data->d_size % nr_maps) != 0) {
 657                 pr_warning("unable to determine map definition size "
 658                            "section %s, %d maps in %zd bytes\n",
 659                            obj->path, nr_maps, data->d_size);
 660                 return -EINVAL;
 661         }
 662
 663         obj->maps = calloc(nr_maps, sizeof(obj->maps[0]));
 664         if (!obj->maps) {
 665                 pr_warning("alloc maps for object failed\n");
 666                 return -ENOMEM;
 667         }
 668         obj->nr_maps = nr_maps;
 669
 670         /*
 671          * fill all fd with -1 so won't close incorrect
 672          * fd (fd=0 is stdin) when failure (zclose won't close
 673          * negative fd)).
 674          */
 675         for (i = 0; i < nr_maps; i++)
 676                 obj->maps[i].fd = -1;
 677
 678         /*
 679          * Fill obj->maps using data in "maps" section.
 680          */
 681         for (i = 0, map_idx = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
 682                 GElf_Sym sym;
 683                 const char *map_name;
 684                 struct bpf_map_def *def;
 685
 686                 if (!gelf_getsym(symbols, i, &sym))
 687                         continue;
 688                 if (sym.st_shndx != obj->efile.maps_shndx)
 689                         continue;
 690
 691                 map_name = elf_strptr(obj->efile.elf,
 692                                       obj->efile.strtabidx,
 693                                       sym.st_name);
 694                 obj->maps[map_idx].offset = sym.st_value;
 695                 if (sym.st_value + map_def_sz > data->d_size) {
 696                         pr_warning("corrupted maps section in %s: last map \"%s\" too small\n",
 697                                    obj->path, map_name);
 698                         return -EINVAL;
 699                 }
 700
 701                 obj->maps[map_idx].name = strdup(map_name);
 702                 if (!obj->maps[map_idx].name) {
 703                         pr_warning("failed to alloc map name\n");
 704                         return -ENOMEM;
 705                 }
 706                 pr_debug("map %d is \"%s\"\n", map_idx,
 707                          obj->maps[map_idx].name);
 708                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
 709                 /*
 710                  * If the definition of the map in the object file fits in
 711                  * bpf_map_def, copy it.  Any extra fields in our version
 712                  * of bpf_map_def will default to zero as a result of the
 713                  * calloc above.
 714                  */
 715                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
 716                         memcpy(&obj->maps[map_idx].def, def, map_def_sz);
 717                 } else {
 718                         /*
 719                          * Here the map structure being read is bigger than what
 720                          * we expect, truncate if the excess bits are all zero.
 721                          * If they are not zero, reject this map as
 722                          * incompatible.
 723                          */
 724                         char *b;
 725                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
 726                              b < ((char *)def) + map_def_sz; b++) {
 727                                 if (*b != 0) {
 728                                         pr_warning("maps section in %s: \"%s\" "
 729                                                    "has unrecognized, non-zero "
 730                                                    "options\n",
 731                                                    obj->path, map_name);
 732                                         return -EINVAL;
 733                                 }
 734                         }
 735                         memcpy(&obj->maps[map_idx].def, def,
 736                                sizeof(struct bpf_map_def));
 737                 }
 738                 map_idx++;
 739         }
 740
 741         qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]), compare_bpf_map);
 742         return 0;
 743 }
 744
 745 static bool section_have_execinstr(struct bpf_object *obj, int idx)
 746 {
 747         Elf_Scn *scn;
 748         GElf_Shdr sh;
 749
 750         scn = elf_getscn(obj->efile.elf, idx);
 751         if (!scn)
 752                 return false;
 753
 754         if (gelf_getshdr(scn, &sh) != &sh)
 755                 return false;
 756
 757         if (sh.sh_flags & SHF_EXECINSTR)
 758                 return true;
 759
 760         return false;
 761 }
 762
 763 static int bpf_object__elf_collect(struct bpf_object *obj)
 764 {
 765         Elf *elf = obj->efile.elf;
 766         GElf_Ehdr *ep = &obj->efile.ehdr;
 767         Elf_Scn *scn = NULL;
 768         int idx = 0, err = 0;
 769
 770         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
 771         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
 772                 pr_warning("failed to get e_shstrndx from %s\n",
 773                            obj->path);
 774                 return -LIBBPF_ERRNO__FORMAT;
 775         }
 776
 777         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 778                 char *name;
 779                 GElf_Shdr sh;
 780                 Elf_Data *data;
 781
 782                 idx++;
 783                 if (gelf_getshdr(scn, &sh) != &sh) {
 784                         pr_warning("failed to get section(%d) header from %s\n",
 785                                    idx, obj->path);
 786                         err = -LIBBPF_ERRNO__FORMAT;
 787                         goto out;
 788                 }
 789
 790                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 791                 if (!name) {
 792                         pr_warning("failed to get section(%d) name from %s\n",
 793                                    idx, obj->path);
 794                         err = -LIBBPF_ERRNO__FORMAT;
 795                         goto out;
 796                 }
 797
 798                 data = elf_getdata(scn, 0);
 799                 if (!data) {
 800                         pr_warning("failed to get section(%d) data from %s(%s)\n",
 801                                    idx, name, obj->path);
 802                         err = -LIBBPF_ERRNO__FORMAT;
 803                         goto out;
 804                 }
 805                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
 806                          idx, name, (unsigned long)data->d_size,
 807                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
 808                          (int)sh.sh_type);
 809
 810                 if (strcmp(name, "license") == 0)
 811                         err = bpf_object__init_license(obj,
 812                                                        data->d_buf,
 813                                                        data->d_size);
 814                 else if (strcmp(name, "version") == 0)
 815                         err = bpf_object__init_kversion(obj,
 816                                                         data->d_buf,
 817                                                         data->d_size);
 818                 else if (strcmp(name, "maps") == 0)
 819                         obj->efile.maps_shndx = idx;
 820                 else if (sh.sh_type == SHT_SYMTAB) {
 821                         if (obj->efile.symbols) {
 822                                 pr_warning("bpf: multiple SYMTAB in %s\n",
 823                                            obj->path);
 824                                 err = -LIBBPF_ERRNO__FORMAT;
 825                         } else {
 826                                 obj->efile.symbols = data;
 827                                 obj->efile.strtabidx = sh.sh_link;
 828                         }
 829                 } else if ((sh.sh_type == SHT_PROGBITS) &&
 830                            (sh.sh_flags & SHF_EXECINSTR) &&
 831                            (data->d_size > 0)) {
 832                         if (strcmp(name, ".text") == 0)
 833                                 obj->efile.text_shndx = idx;
 834                         err = bpf_object__add_program(obj, data->d_buf,
 835                                                       data->d_size, name, idx);
 836                         if (err) {
 837                                 char errmsg[STRERR_BUFSIZE];
 838
 839                                 strerror_r(-err, errmsg, sizeof(errmsg));
 840                                 pr_warning("failed to alloc program %s (%s): %s",
 841                                            name, obj->path, errmsg);
 842                         }
 843                 } else if (sh.sh_type == SHT_REL) {
 844                         void *reloc = obj->efile.reloc;
 845                         int nr_reloc = obj->efile.nr_reloc + 1;
 846                         int sec = sh.sh_info; /* points to other section */
 847
 848                         /* Only do relo for section with exec instructions */
 849                         if (!section_have_execinstr(obj, sec)) {
 850                                 pr_debug("skip relo %s(%d) for section(%d)\n",
 851                                          name, idx, sec);
 852                                 continue;
 853                         }
 854
 855                         reloc = realloc(reloc,
 856                                         sizeof(*obj->efile.reloc) * nr_reloc);
 857                         if (!reloc) {
 858                                 pr_warning("realloc failed\n");
 859                                 err = -ENOMEM;
 860                         } else {
 861                                 int n = nr_reloc - 1;
 862
 863                                 obj->efile.reloc = reloc;
 864                                 obj->efile.nr_reloc = nr_reloc;
 865
 866                                 obj->efile.reloc[n].shdr = sh;
 867                                 obj->efile.reloc[n].data = data;
 868                         }
 869                 } else {
 870                         pr_debug("skip section(%d) %s\n", idx, name);
 871                 }
 872                 if (err)
 873                         goto out;
 874         }
 875
 876         if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
 877                 pr_warning("Corrupted ELF file: index of strtab invalid\n");
 878                 return LIBBPF_ERRNO__FORMAT;
 879         }
 880         if (obj->efile.maps_shndx >= 0) {
 881                 err = bpf_object__init_maps(obj);
 882                 if (err)
 883                         goto out;
 884         }
 885         err = bpf_object__init_prog_names(obj);
 886 out:
 887         return err;
 888 }
 889
 890 static struct bpf_program *
 891 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
 892 {
 893         struct bpf_program *prog;
 894         size_t i;
 895
 896         for (i = 0; i < obj->nr_programs; i++) {
 897                 prog = &obj->programs[i];
 898                 if (prog->idx == idx)
 899                         return prog;
 900         }
 901         return NULL;
 902 }
 903
 904 static int
 905 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
 906                            Elf_Data *data, struct bpf_object *obj)
 907 {
 908         Elf_Data *symbols = obj->efile.symbols;
 909         int text_shndx = obj->efile.text_shndx;
 910         int maps_shndx = obj->efile.maps_shndx;
 911         struct bpf_map *maps = obj->maps;
 912         size_t nr_maps = obj->nr_maps;
 913         int i, nrels;
 914
 915         pr_debug("collecting relocating info for: '%s'\n",
 916                  prog->section_name);
 917         nrels = shdr->sh_size / shdr->sh_entsize;
 918
 919         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
 920         if (!prog->reloc_desc) {
 921                 pr_warning("failed to alloc memory in relocation\n");
 922                 return -ENOMEM;
 923         }
 924         prog->nr_reloc = nrels;
 925
 926         for (i = 0; i < nrels; i++) {
 927                 GElf_Sym sym;
 928                 GElf_Rel rel;
 929                 unsigned int insn_idx;
 930                 struct bpf_insn *insns = prog->insns;
 931                 size_t map_idx;
 932
 933                 if (!gelf_getrel(data, i, &rel)) {
 934                         pr_warning("relocation: failed to get %d reloc\n", i);
 935                         return -LIBBPF_ERRNO__FORMAT;
 936                 }
 937
 938                 if (!gelf_getsym(symbols,
 939                                  GELF_R_SYM(rel.r_info),
 940                                  &sym)) {
 941                         pr_warning("relocation: symbol %"PRIx64" not found\n",
 942                                    GELF_R_SYM(rel.r_info));
 943                         return -LIBBPF_ERRNO__FORMAT;
 944                 }
 945                 pr_debug("relo for %lld value %lld name %d\n",
 946                          (long long) (rel.r_info >> 32),
 947                          (long long) sym.st_value, sym.st_name);
 948
 949                 if (sym.st_shndx != maps_shndx && sym.st_shndx != text_shndx) {
 950                         pr_warning("Program '%s' contains non-map related relo data pointing to section %u\n",
 951                                    prog->section_name, sym.st_shndx);
 952                         return -LIBBPF_ERRNO__RELOC;
 953                 }
 954
 955                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
 956                 pr_debug("relocation: insn_idx=%u\n", insn_idx);
 957
 958                 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
 959                         if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
 960                                 pr_warning("incorrect bpf_call opcode\n");
 961                                 return -LIBBPF_ERRNO__RELOC;
 962                         }
 963                         prog->reloc_desc[i].type = RELO_CALL;
 964                         prog->reloc_desc[i].insn_idx = insn_idx;
 965                         prog->reloc_desc[i].text_off = sym.st_value;
 966                         continue;
 967                 }
 968
 969                 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
 970                         pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
 971                                    insn_idx, insns[insn_idx].code);
 972                         return -LIBBPF_ERRNO__RELOC;
 973                 }
 974
 975                 /* TODO: 'maps' is sorted. We can use bsearch to make it faster. */
 976                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
 977                         if (maps[map_idx].offset == sym.st_value) {
 978                                 pr_debug("relocation: find map %zd (%s) for insn %u\n",
 979                                          map_idx, maps[map_idx].name, insn_idx);
 980                                 break;
 981                         }
 982                 }
 983
 984                 if (map_idx >= nr_maps) {
 985                         pr_warning("bpf relocation: map_idx %d large than %d\n",
 986                                    (int)map_idx, (int)nr_maps - 1);
 987                         return -LIBBPF_ERRNO__RELOC;
 988                 }
 989
 990                 prog->reloc_desc[i].type = RELO_LD64;
 991                 prog->reloc_desc[i].insn_idx = insn_idx;
 992                 prog->reloc_desc[i].map_idx = map_idx;
 993         }
 994         return 0;
 995 }
 996
 997 static int
 998 bpf_object__create_maps(struct bpf_object *obj)
 999 {
1000         unsigned int i;
1001
1002         for (i = 0; i < obj->nr_maps; i++) {
1003                 struct bpf_map_def *def = &obj->maps[i].def;
1004                 int *pfd = &obj->maps[i].fd;
1005
1006                 *pfd = bpf_create_map_name(def->type,
1007                                            obj->maps[i].name,
1008                                            def->key_size,
1009                                            def->value_size,
1010                                            def->max_entries,
1011                                            def->map_flags);
1012                 if (*pfd < 0) {
1013                         size_t j;
1014                         int err = *pfd;
1015
1016                         pr_warning("failed to create map (name: '%s'): %s\n",
1017                                    obj->maps[i].name,
1018                                    strerror(errno));
1019                         for (j = 0; j < i; j++)
1020                                 zclose(obj->maps[j].fd);
1021                         return err;
1022                 }
1023                 pr_debug("create map %s: fd=%d\n", obj->maps[i].name, *pfd);
1024         }
1025
1026         return 0;
1027 }
1028
1029 static int
1030 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
1031                         struct reloc_desc *relo)
1032 {
1033         struct bpf_insn *insn, *new_insn;
1034         struct bpf_program *text;
1035         size_t new_cnt;
1036
1037         if (relo->type != RELO_CALL)
1038                 return -LIBBPF_ERRNO__RELOC;
1039
1040         if (prog->idx == obj->efile.text_shndx) {
1041                 pr_warning("relo in .text insn %d into off %d\n",
1042                            relo->insn_idx, relo->text_off);
1043                 return -LIBBPF_ERRNO__RELOC;
1044         }
1045
1046         if (prog->main_prog_cnt == 0) {
1047                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
1048                 if (!text) {
1049                         pr_warning("no .text section found yet relo into text exist\n");
1050                         return -LIBBPF_ERRNO__RELOC;
1051                 }
1052                 new_cnt = prog->insns_cnt + text->insns_cnt;
1053                 new_insn = realloc(prog->insns, new_cnt * sizeof(*insn));
1054                 if (!new_insn) {
1055                         pr_warning("oom in prog realloc\n");
1056                         return -ENOMEM;
1057                 }
1058                 memcpy(new_insn + prog->insns_cnt, text->insns,
1059                        text->insns_cnt * sizeof(*insn));
1060                 prog->insns = new_insn;
1061                 prog->main_prog_cnt = prog->insns_cnt;
1062                 prog->insns_cnt = new_cnt;
1063                 pr_debug("added %zd insn from %s to prog %s\n",
1064                          text->insns_cnt, text->section_name,
1065                          prog->section_name);
1066         }
1067         insn = &prog->insns[relo->insn_idx];
1068         insn->imm += prog->main_prog_cnt - relo->insn_idx;
1069         return 0;
1070 }
1071
1072 static int
1073 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
1074 {
1075         int i, err;
1076
1077         if (!prog || !prog->reloc_desc)
1078                 return 0;
1079
1080         for (i = 0; i < prog->nr_reloc; i++) {
1081                 if (prog->reloc_desc[i].type == RELO_LD64) {
1082                         struct bpf_insn *insns = prog->insns;
1083                         int insn_idx, map_idx;
1084
1085                         insn_idx = prog->reloc_desc[i].insn_idx;
1086                         map_idx = prog->reloc_desc[i].map_idx;
1087
1088                         if (insn_idx >= (int)prog->insns_cnt) {
1089                                 pr_warning("relocation out of range: '%s'\n",
1090                                            prog->section_name);
1091                                 return -LIBBPF_ERRNO__RELOC;
1092                         }
1093                         insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
1094                         insns[insn_idx].imm = obj->maps[map_idx].fd;
1095                 } else {
1096                         err = bpf_program__reloc_text(prog, obj,
1097                                                       &prog->reloc_desc[i]);
1098                         if (err)
1099                                 return err;
1100                 }
1101         }
1102
1103         zfree(&prog->reloc_desc);
1104         prog->nr_reloc = 0;
1105         return 0;
1106 }
1107
1108
1109 static int
1110 bpf_object__relocate(struct bpf_object *obj)
1111 {
1112         struct bpf_program *prog;
1113         size_t i;
1114         int err;
1115
1116         for (i = 0; i < obj->nr_programs; i++) {
1117                 prog = &obj->programs[i];
1118
1119                 err = bpf_program__relocate(prog, obj);
1120                 if (err) {
1121                         pr_warning("failed to relocate '%s'\n",
1122                                    prog->section_name);
1123                         return err;
1124                 }
1125         }
1126         return 0;
1127 }
1128
1129 static int bpf_object__collect_reloc(struct bpf_object *obj)
1130 {
1131         int i, err;
1132
1133         if (!obj_elf_valid(obj)) {
1134                 pr_warning("Internal error: elf object is closed\n");
1135                 return -LIBBPF_ERRNO__INTERNAL;
1136         }
1137
1138         for (i = 0; i < obj->efile.nr_reloc; i++) {
1139                 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
1140                 Elf_Data *data = obj->efile.reloc[i].data;
1141                 int idx = shdr->sh_info;
1142                 struct bpf_program *prog;
1143
1144                 if (shdr->sh_type != SHT_REL) {
1145                         pr_warning("internal error at %d\n", __LINE__);
1146                         return -LIBBPF_ERRNO__INTERNAL;
1147                 }
1148
1149                 prog = bpf_object__find_prog_by_idx(obj, idx);
1150                 if (!prog) {
1151                         pr_warning("relocation failed: no section(%d)\n", idx);
1152                         return -LIBBPF_ERRNO__RELOC;
1153                 }
1154
1155                 err = bpf_program__collect_reloc(prog,
1156                                                  shdr, data,
1157                                                  obj);
1158                 if (err)
1159                         return err;
1160         }
1161         return 0;
1162 }
1163
1164 static int
1165 load_program(enum bpf_prog_type type, const char *name, struct bpf_insn *insns,
1166              int insns_cnt, char *license, u32 kern_version, int *pfd)
1167 {
1168         int ret;
1169         char *log_buf;
1170
1171         if (!insns || !insns_cnt)
1172                 return -EINVAL;
1173
1174         log_buf = malloc(BPF_LOG_BUF_SIZE);
1175         if (!log_buf)
1176                 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
1177
1178         ret = bpf_load_program_name(type, name, insns, insns_cnt, license,
1179                                     kern_version, log_buf, BPF_LOG_BUF_SIZE);
1180
1181         if (ret >= 0) {
1182                 *pfd = ret;
1183                 ret = 0;
1184                 goto out;
1185         }
1186
1187         ret = -LIBBPF_ERRNO__LOAD;
1188         pr_warning("load bpf program failed: %s\n", strerror(errno));
1189
1190         if (log_buf && log_buf[0] != '\0') {
1191                 ret = -LIBBPF_ERRNO__VERIFY;
1192                 pr_warning("-- BEGIN DUMP LOG ---\n");
1193                 pr_warning("\n%s\n", log_buf);
1194                 pr_warning("-- END LOG --\n");
1195         } else if (insns_cnt >= BPF_MAXINSNS) {
1196                 pr_warning("Program too large (%d insns), at most %d insns\n",
1197                            insns_cnt, BPF_MAXINSNS);
1198                 ret = -LIBBPF_ERRNO__PROG2BIG;
1199         } else {
1200                 /* Wrong program type? */
1201                 if (type != BPF_PROG_TYPE_KPROBE) {
1202                         int fd;
1203
1204                         fd = bpf_load_program_name(BPF_PROG_TYPE_KPROBE, name,
1205                                                    insns, insns_cnt, license,
1206                                                    kern_version, NULL, 0);
1207                         if (fd >= 0) {
1208                                 close(fd);
1209                                 ret = -LIBBPF_ERRNO__PROGTYPE;
1210                                 goto out;
1211                         }
1212                 }
1213
1214                 if (log_buf)
1215                         ret = -LIBBPF_ERRNO__KVER;
1216         }
1217
1218 out:
1219         free(log_buf);
1220         return ret;
1221 }
1222
1223 static int
1224 bpf_program__load(struct bpf_program *prog,
1225                   char *license, u32 kern_version)
1226 {
1227         int err = 0, fd, i;
1228
1229         if (prog->instances.nr < 0 || !prog->instances.fds) {
1230                 if (prog->preprocessor) {
1231                         pr_warning("Internal error: can't load program '%s'\n",
1232                                    prog->section_name);
1233                         return -LIBBPF_ERRNO__INTERNAL;
1234                 }
1235
1236                 prog->instances.fds = malloc(sizeof(int));
1237                 if (!prog->instances.fds) {
1238                         pr_warning("Not enough memory for BPF fds\n");
1239                         return -ENOMEM;
1240                 }
1241                 prog->instances.nr = 1;
1242                 prog->instances.fds[0] = -1;
1243         }
1244
1245         if (!prog->preprocessor) {
1246                 if (prog->instances.nr != 1) {
1247                         pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
1248                                    prog->section_name, prog->instances.nr);
1249                 }
1250                 err = load_program(prog->type, prog->name, prog->insns,
1251                                    prog->insns_cnt, license, kern_version, &fd);
1252                 if (!err)
1253                         prog->instances.fds[0] = fd;
1254                 goto out;
1255         }
1256
1257         for (i = 0; i < prog->instances.nr; i++) {
1258                 struct bpf_prog_prep_result result;
1259                 bpf_program_prep_t preprocessor = prog->preprocessor;
1260
1261                 bzero(&result, sizeof(result));
1262                 err = preprocessor(prog, i, prog->insns,
1263                                    prog->insns_cnt, &result);
1264                 if (err) {
1265                         pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
1266                                    i, prog->section_name);
1267                         goto out;
1268                 }
1269
1270                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
1271                         pr_debug("Skip loading the %dth instance of program '%s'\n",
1272                                  i, prog->section_name);
1273                         prog->instances.fds[i] = -1;
1274                         if (result.pfd)
1275                                 *result.pfd = -1;
1276                         continue;
1277                 }
1278
1279                 err = load_program(prog->type, prog->name,
1280                                    result.new_insn_ptr,
1281                                    result.new_insn_cnt,
1282                                    license, kern_version, &fd);
1283
1284                 if (err) {
1285                         pr_warning("Loading the %dth instance of program '%s' failed\n",
1286                                         i, prog->section_name);
1287                         goto out;
1288                 }
1289
1290                 if (result.pfd)
1291                         *result.pfd = fd;
1292                 prog->instances.fds[i] = fd;
1293         }
1294 out:
1295         if (err)
1296                 pr_warning("failed to load program '%s'\n",
1297                            prog->section_name);
1298         zfree(&prog->insns);
1299         prog->insns_cnt = 0;
1300         return err;
1301 }
1302
1303 static int
1304 bpf_object__load_progs(struct bpf_object *obj)
1305 {
1306         size_t i;
1307         int err;
1308
1309         for (i = 0; i < obj->nr_programs; i++) {
1310                 if (obj->programs[i].idx == obj->efile.text_shndx)
1311                         continue;
1312                 err = bpf_program__load(&obj->programs[i],
1313                                         obj->license,
1314                                         obj->kern_version);
1315                 if (err)
1316                         return err;
1317         }
1318         return 0;
1319 }
1320
1321 static int bpf_object__validate(struct bpf_object *obj)
1322 {
1323         if (obj->kern_version == 0) {
1324                 pr_warning("%s doesn't provide kernel version\n",
1325                            obj->path);
1326                 return -LIBBPF_ERRNO__KVERSION;
1327         }
1328         return 0;
1329 }
1330
1331 static struct bpf_object *
1332 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz)
1333 {
1334         struct bpf_object *obj;
1335         int err;
1336
1337         if (elf_version(EV_CURRENT) == EV_NONE) {
1338                 pr_warning("failed to init libelf for %s\n", path);
1339                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1340         }
1341
1342         obj = bpf_object__new(path, obj_buf, obj_buf_sz);
1343         if (IS_ERR(obj))
1344                 return obj;
1345
1346         CHECK_ERR(bpf_object__elf_init(obj), err, out);
1347         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
1348         CHECK_ERR(bpf_object__elf_collect(obj), err, out);
1349         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
1350         CHECK_ERR(bpf_object__validate(obj), err, out);
1351
1352         bpf_object__elf_finish(obj);
1353         return obj;
1354 out:
1355         bpf_object__close(obj);
1356         return ERR_PTR(err);
1357 }
1358
1359 struct bpf_object *bpf_object__open(const char *path)
1360 {
1361         /* param validation */
1362         if (!path)
1363                 return NULL;
1364
1365         pr_debug("loading %s\n", path);
1366
1367         return __bpf_object__open(path, NULL, 0);
1368 }
1369
1370 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
1371                                            size_t obj_buf_sz,
1372                                            const char *name)
1373 {
1374         char tmp_name[64];
1375
1376         /* param validation */
1377         if (!obj_buf || obj_buf_sz <= 0)
1378                 return NULL;
1379
1380         if (!name) {
1381                 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
1382                          (unsigned long)obj_buf,
1383                          (unsigned long)obj_buf_sz);
1384                 tmp_name[sizeof(tmp_name) - 1] = '\0';
1385                 name = tmp_name;
1386         }
1387         pr_debug("loading object '%s' from buffer\n",
1388                  name);
1389
1390         return __bpf_object__open(name, obj_buf, obj_buf_sz);
1391 }
1392
1393 int bpf_object__unload(struct bpf_object *obj)
1394 {
1395         size_t i;
1396
1397         if (!obj)
1398                 return -EINVAL;
1399
1400         for (i = 0; i < obj->nr_maps; i++)
1401                 zclose(obj->maps[i].fd);
1402
1403         for (i = 0; i < obj->nr_programs; i++)
1404                 bpf_program__unload(&obj->programs[i]);
1405
1406         return 0;
1407 }
1408
1409 int bpf_object__load(struct bpf_object *obj)
1410 {
1411         int err;
1412
1413         if (!obj)
1414                 return -EINVAL;
1415
1416         if (obj->loaded) {
1417                 pr_warning("object should not be loaded twice\n");
1418                 return -EINVAL;
1419         }
1420
1421         obj->loaded = true;
1422
1423         CHECK_ERR(bpf_object__create_maps(obj), err, out);
1424         CHECK_ERR(bpf_object__relocate(obj), err, out);
1425         CHECK_ERR(bpf_object__load_progs(obj), err, out);
1426
1427         return 0;
1428 out:
1429         bpf_object__unload(obj);
1430         pr_warning("failed to load object '%s'\n", obj->path);
1431         return err;
1432 }
1433
1434 static int check_path(const char *path)
1435 {
1436         struct statfs st_fs;
1437         char *dname, *dir;
1438         int err = 0;
1439
1440         if (path == NULL)
1441                 return -EINVAL;
1442
1443         dname = strdup(path);
1444         if (dname == NULL)
1445                 return -ENOMEM;
1446
1447         dir = dirname(dname);
1448         if (statfs(dir, &st_fs)) {
1449                 pr_warning("failed to statfs %s: %s\n", dir, strerror(errno));
1450                 err = -errno;
1451         }
1452         free(dname);
1453
1454         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
1455                 pr_warning("specified path %s is not on BPF FS\n", path);
1456                 err = -EINVAL;
1457         }
1458
1459         return err;
1460 }
1461
1462 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
1463                               int instance)
1464 {
1465         int err;
1466
1467         err = check_path(path);
1468         if (err)
1469                 return err;
1470
1471         if (prog == NULL) {
1472                 pr_warning("invalid program pointer\n");
1473                 return -EINVAL;
1474         }
1475
1476         if (instance < 0 || instance >= prog->instances.nr) {
1477                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
1478                            instance, prog->section_name, prog->instances.nr);
1479                 return -EINVAL;
1480         }
1481
1482         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
1483                 pr_warning("failed to pin program: %s\n", strerror(errno));
1484                 return -errno;
1485         }
1486         pr_debug("pinned program '%s'\n", path);
1487
1488         return 0;
1489 }
1490
1491 static int make_dir(const char *path)
1492 {
1493         int err = 0;
1494
1495         if (mkdir(path, 0700) && errno != EEXIST)
1496                 err = -errno;
1497
1498         if (err)
1499                 pr_warning("failed to mkdir %s: %s\n", path, strerror(-err));
1500         return err;
1501 }
1502
1503 int bpf_program__pin(struct bpf_program *prog, const char *path)
1504 {
1505         int i, err;
1506
1507         err = check_path(path);
1508         if (err)
1509                 return err;
1510
1511         if (prog == NULL) {
1512                 pr_warning("invalid program pointer\n");
1513                 return -EINVAL;
1514         }
1515
1516         if (prog->instances.nr <= 0) {
1517                 pr_warning("no instances of prog %s to pin\n",
1518                            prog->section_name);
1519                 return -EINVAL;
1520         }
1521
1522         err = make_dir(path);
1523         if (err)
1524                 return err;
1525
1526         for (i = 0; i < prog->instances.nr; i++) {
1527                 char buf[PATH_MAX];
1528                 int len;
1529
1530                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
1531                 if (len < 0)
1532                         return -EINVAL;
1533                 else if (len >= PATH_MAX)
1534                         return -ENAMETOOLONG;
1535
1536                 err = bpf_program__pin_instance(prog, buf, i);
1537                 if (err)
1538                         return err;
1539         }
1540
1541         return 0;
1542 }
1543
1544 int bpf_map__pin(struct bpf_map *map, const char *path)
1545 {
1546         int err;
1547
1548         err = check_path(path);
1549         if (err)
1550                 return err;
1551
1552         if (map == NULL) {
1553                 pr_warning("invalid map pointer\n");
1554                 return -EINVAL;
1555         }
1556
1557         if (bpf_obj_pin(map->fd, path)) {
1558                 pr_warning("failed to pin map: %s\n", strerror(errno));
1559                 return -errno;
1560         }
1561
1562         pr_debug("pinned map '%s'\n", path);
1563         return 0;
1564 }
1565
1566 int bpf_object__pin(struct bpf_object *obj, const char *path)
1567 {
1568         struct bpf_program *prog;
1569         struct bpf_map *map;
1570         int err;
1571
1572         if (!obj)
1573                 return -ENOENT;
1574
1575         if (!obj->loaded) {
1576                 pr_warning("object not yet loaded; load it first\n");
1577                 return -ENOENT;
1578         }
1579
1580         err = make_dir(path);
1581         if (err)
1582                 return err;
1583
1584         bpf_map__for_each(map, obj) {
1585                 char buf[PATH_MAX];
1586                 int len;
1587
1588                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1589                                bpf_map__name(map));
1590                 if (len < 0)
1591                         return -EINVAL;
1592                 else if (len >= PATH_MAX)
1593                         return -ENAMETOOLONG;
1594
1595                 err = bpf_map__pin(map, buf);
1596                 if (err)
1597                         return err;
1598         }
1599
1600         bpf_object__for_each_program(prog, obj) {
1601                 char buf[PATH_MAX];
1602                 int len;
1603
1604                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1605                                prog->section_name);
1606                 if (len < 0)
1607                         return -EINVAL;
1608                 else if (len >= PATH_MAX)
1609                         return -ENAMETOOLONG;
1610
1611                 err = bpf_program__pin(prog, buf);
1612                 if (err)
1613                         return err;
1614         }
1615
1616         return 0;
1617 }
1618
1619 void bpf_object__close(struct bpf_object *obj)
1620 {
1621         size_t i;
1622
1623         if (!obj)
1624                 return;
1625
1626         if (obj->clear_priv)
1627                 obj->clear_priv(obj, obj->priv);
1628
1629         bpf_object__elf_finish(obj);
1630         bpf_object__unload(obj);
1631
1632         for (i = 0; i < obj->nr_maps; i++) {
1633                 zfree(&obj->maps[i].name);
1634                 if (obj->maps[i].clear_priv)
1635                         obj->maps[i].clear_priv(&obj->maps[i],
1636                                                 obj->maps[i].priv);
1637                 obj->maps[i].priv = NULL;
1638                 obj->maps[i].clear_priv = NULL;
1639         }
1640         zfree(&obj->maps);
1641         obj->nr_maps = 0;
1642
1643         if (obj->programs && obj->nr_programs) {
1644                 for (i = 0; i < obj->nr_programs; i++)
1645                         bpf_program__exit(&obj->programs[i]);
1646         }
1647         zfree(&obj->programs);
1648
1649         list_del(&obj->list);
1650         free(obj);
1651 }
1652
1653 struct bpf_object *
1654 bpf_object__next(struct bpf_object *prev)
1655 {
1656         struct bpf_object *next;
1657
1658         if (!prev)
1659                 next = list_first_entry(&bpf_objects_list,
1660                                         struct bpf_object,
1661                                         list);
1662         else
1663                 next = list_next_entry(prev, list);
1664
1665         /* Empty list is noticed here so don't need checking on entry. */
1666         if (&next->list == &bpf_objects_list)
1667                 return NULL;
1668
1669         return next;
1670 }
1671
1672 const char *bpf_object__name(struct bpf_object *obj)
1673 {
1674         return obj ? obj->path : ERR_PTR(-EINVAL);
1675 }
1676
1677 unsigned int bpf_object__kversion(struct bpf_object *obj)
1678 {
1679         return obj ? obj->kern_version : 0;
1680 }
1681
1682 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
1683                          bpf_object_clear_priv_t clear_priv)
1684 {
1685         if (obj->priv && obj->clear_priv)
1686                 obj->clear_priv(obj, obj->priv);
1687
1688         obj->priv = priv;
1689         obj->clear_priv = clear_priv;
1690         return 0;
1691 }
1692
1693 void *bpf_object__priv(struct bpf_object *obj)
1694 {
1695         return obj ? obj->priv : ERR_PTR(-EINVAL);
1696 }
1697
1698 struct bpf_program *
1699 bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1700 {
1701         size_t idx;
1702
1703         if (!obj->programs)
1704                 return NULL;
1705         /* First handler */
1706         if (prev == NULL)
1707                 return &obj->programs[0];
1708
1709         if (prev->obj != obj) {
1710                 pr_warning("error: program handler doesn't match object\n");
1711                 return NULL;
1712         }
1713
1714         idx = (prev - obj->programs) + 1;
1715         if (idx >= obj->nr_programs)
1716                 return NULL;
1717         return &obj->programs[idx];
1718 }
1719
1720 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
1721                           bpf_program_clear_priv_t clear_priv)
1722 {
1723         if (prog->priv && prog->clear_priv)
1724                 prog->clear_priv(prog, prog->priv);
1725
1726         prog->priv = priv;
1727         prog->clear_priv = clear_priv;
1728         return 0;
1729 }
1730
1731 void *bpf_program__priv(struct bpf_program *prog)
1732 {
1733         return prog ? prog->priv : ERR_PTR(-EINVAL);
1734 }
1735
1736 const char *bpf_program__title(struct bpf_program *prog, bool needs_copy)
1737 {
1738         const char *title;
1739
1740         title = prog->section_name;
1741         if (needs_copy) {
1742                 title = strdup(title);
1743                 if (!title) {
1744                         pr_warning("failed to strdup program title\n");
1745                         return ERR_PTR(-ENOMEM);
1746                 }
1747         }
1748
1749         return title;
1750 }
1751
1752 int bpf_program__fd(struct bpf_program *prog)
1753 {
1754         return bpf_program__nth_fd(prog, 0);
1755 }
1756
1757 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
1758                           bpf_program_prep_t prep)
1759 {
1760         int *instances_fds;
1761
1762         if (nr_instances <= 0 || !prep)
1763                 return -EINVAL;
1764
1765         if (prog->instances.nr > 0 || prog->instances.fds) {
1766                 pr_warning("Can't set pre-processor after loading\n");
1767                 return -EINVAL;
1768         }
1769
1770         instances_fds = malloc(sizeof(int) * nr_instances);
1771         if (!instances_fds) {
1772                 pr_warning("alloc memory failed for fds\n");
1773                 return -ENOMEM;
1774         }
1775
1776         /* fill all fd with -1 */
1777         memset(instances_fds, -1, sizeof(int) * nr_instances);
1778
1779         prog->instances.nr = nr_instances;
1780         prog->instances.fds = instances_fds;
1781         prog->preprocessor = prep;
1782         return 0;
1783 }
1784
1785 int bpf_program__nth_fd(struct bpf_program *prog, int n)
1786 {
1787         int fd;
1788
1789         if (n >= prog->instances.nr || n < 0) {
1790                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
1791                            n, prog->section_name, prog->instances.nr);
1792                 return -EINVAL;
1793         }
1794
1795         fd = prog->instances.fds[n];
1796         if (fd < 0) {
1797                 pr_warning("%dth instance of program '%s' is invalid\n",
1798                            n, prog->section_name);
1799                 return -ENOENT;
1800         }
1801
1802         return fd;
1803 }
1804
1805 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
1806 {
1807         prog->type = type;
1808 }
1809
1810 static bool bpf_program__is_type(struct bpf_program *prog,
1811                                  enum bpf_prog_type type)
1812 {
1813         return prog ? (prog->type == type) : false;
1814 }
1815
1816 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                   \
1817 int bpf_program__set_##NAME(struct bpf_program *prog)   \
1818 {                                                       \
1819         if (!prog)                                      \
1820                 return -EINVAL;                         \
1821         bpf_program__set_type(prog, TYPE);              \
1822         return 0;                                       \
1823 }                                                       \
1824                                                         \
1825 bool bpf_program__is_##NAME(struct bpf_program *prog)   \
1826 {                                                       \
1827         return bpf_program__is_type(prog, TYPE);        \
1828 }                                                       \
1829
1830 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
1831 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
1832 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
1833 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
1834 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
1835 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
1836 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
1837
1838 #define BPF_PROG_SEC(string, type) { string, sizeof(string) - 1, type }
1839 static const struct {
1840         const char *sec;
1841         size_t len;
1842         enum bpf_prog_type prog_type;
1843 } section_names[] = {
1844         BPF_PROG_SEC("socket",          BPF_PROG_TYPE_SOCKET_FILTER),
1845         BPF_PROG_SEC("kprobe/",         BPF_PROG_TYPE_KPROBE),
1846         BPF_PROG_SEC("kretprobe/",      BPF_PROG_TYPE_KPROBE),
1847         BPF_PROG_SEC("classifier",      BPF_PROG_TYPE_SCHED_CLS),
1848         BPF_PROG_SEC("action",          BPF_PROG_TYPE_SCHED_ACT),
1849         BPF_PROG_SEC("tracepoint/",     BPF_PROG_TYPE_TRACEPOINT),
1850         BPF_PROG_SEC("xdp",             BPF_PROG_TYPE_XDP),
1851         BPF_PROG_SEC("perf_event",      BPF_PROG_TYPE_PERF_EVENT),
1852         BPF_PROG_SEC("cgroup/skb",      BPF_PROG_TYPE_CGROUP_SKB),
1853         BPF_PROG_SEC("cgroup/sock",     BPF_PROG_TYPE_CGROUP_SOCK),
1854         BPF_PROG_SEC("cgroup/dev",      BPF_PROG_TYPE_CGROUP_DEVICE),
1855         BPF_PROG_SEC("lwt_in",          BPF_PROG_TYPE_LWT_IN),
1856         BPF_PROG_SEC("lwt_out",         BPF_PROG_TYPE_LWT_OUT),
1857         BPF_PROG_SEC("lwt_xmit",        BPF_PROG_TYPE_LWT_XMIT),
1858         BPF_PROG_SEC("sockops",         BPF_PROG_TYPE_SOCK_OPS),
1859         BPF_PROG_SEC("sk_skb",          BPF_PROG_TYPE_SK_SKB),
1860 };
1861 #undef BPF_PROG_SEC
1862
1863 static enum bpf_prog_type bpf_program__guess_type(struct bpf_program *prog)
1864 {
1865         int i;
1866
1867         if (!prog->section_name)
1868                 goto err;
1869
1870         for (i = 0; i < ARRAY_SIZE(section_names); i++)
1871                 if (strncmp(prog->section_name, section_names[i].sec,
1872                             section_names[i].len) == 0)
1873                         return section_names[i].prog_type;
1874
1875 err:
1876         pr_warning("failed to guess program type based on section name %s\n",
1877                    prog->section_name);
1878
1879         return BPF_PROG_TYPE_UNSPEC;
1880 }
1881
1882 int bpf_map__fd(struct bpf_map *map)
1883 {
1884         return map ? map->fd : -EINVAL;
1885 }
1886
1887 const struct bpf_map_def *bpf_map__def(struct bpf_map *map)
1888 {
1889         return map ? &map->def : ERR_PTR(-EINVAL);
1890 }
1891
1892 const char *bpf_map__name(struct bpf_map *map)
1893 {
1894         return map ? map->name : NULL;
1895 }
1896
1897 int bpf_map__set_priv(struct bpf_map *map, void *priv,
1898                      bpf_map_clear_priv_t clear_priv)
1899 {
1900         if (!map)
1901                 return -EINVAL;
1902
1903         if (map->priv) {
1904                 if (map->clear_priv)
1905                         map->clear_priv(map, map->priv);
1906         }
1907
1908         map->priv = priv;
1909         map->clear_priv = clear_priv;
1910         return 0;
1911 }
1912
1913 void *bpf_map__priv(struct bpf_map *map)
1914 {
1915         return map ? map->priv : ERR_PTR(-EINVAL);
1916 }
1917
1918 struct bpf_map *
1919 bpf_map__next(struct bpf_map *prev, struct bpf_object *obj)
1920 {
1921         size_t idx;
1922         struct bpf_map *s, *e;
1923
1924         if (!obj || !obj->maps)
1925                 return NULL;
1926
1927         s = obj->maps;
1928         e = obj->maps + obj->nr_maps;
1929
1930         if (prev == NULL)
1931                 return s;
1932
1933         if ((prev < s) || (prev >= e)) {
1934                 pr_warning("error in %s: map handler doesn't belong to object\n",
1935                            __func__);
1936                 return NULL;
1937         }
1938
1939         idx = (prev - obj->maps) + 1;
1940         if (idx >= obj->nr_maps)
1941                 return NULL;
1942         return &obj->maps[idx];
1943 }
1944
1945 struct bpf_map *
1946 bpf_object__find_map_by_name(struct bpf_object *obj, const char *name)
1947 {
1948         struct bpf_map *pos;
1949
1950         bpf_map__for_each(pos, obj) {
1951                 if (pos->name && !strcmp(pos->name, name))
1952                         return pos;
1953         }
1954         return NULL;
1955 }
1956
1957 struct bpf_map *
1958 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
1959 {
1960         int i;
1961
1962         for (i = 0; i < obj->nr_maps; i++) {
1963                 if (obj->maps[i].offset == offset)
1964                         return &obj->maps[i];
1965         }
1966         return ERR_PTR(-ENOENT);
1967 }
1968
1969 long libbpf_get_error(const void *ptr)
1970 {
1971         if (IS_ERR(ptr))
1972                 return PTR_ERR(ptr);
1973         return 0;
1974 }
1975
1976 int bpf_prog_load(const char *file, enum bpf_prog_type type,
1977                   struct bpf_object **pobj, int *prog_fd)
1978 {
1979         struct bpf_program *prog, *first_prog = NULL;
1980         struct bpf_object *obj;
1981         int err;
1982
1983         obj = bpf_object__open(file);
1984         if (IS_ERR(obj))
1985                 return -ENOENT;
1986
1987         bpf_object__for_each_program(prog, obj) {
1988                 /*
1989                  * If type is not specified, try to guess it based on
1990                  * section name.
1991                  */
1992                 if (type == BPF_PROG_TYPE_UNSPEC) {
1993                         type = bpf_program__guess_type(prog);
1994                         if (type == BPF_PROG_TYPE_UNSPEC) {
1995                                 bpf_object__close(obj);
1996                                 return -EINVAL;
1997                         }
1998                 }
1999
2000                 bpf_program__set_type(prog, type);
2001                 if (prog->idx != obj->efile.text_shndx && !first_prog)
2002                         first_prog = prog;
2003         }
2004
2005         if (!first_prog) {
2006                 pr_warning("object file doesn't contain bpf program\n");
2007                 bpf_object__close(obj);
2008                 return -ENOENT;
2009         }
2010
2011         err = bpf_object__load(obj);
2012         if (err) {
2013                 bpf_object__close(obj);
2014                 return -EINVAL;
2015         }
2016
2017         *pobj = obj;
2018         *prog_fd = bpf_program__fd(first_prog);
2019         return 0;
2020 }