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 int bpf_object__elf_collect(struct bpf_object *obj)
 746 {
 747         Elf *elf = obj->efile.elf;
 748         GElf_Ehdr *ep = &obj->efile.ehdr;
 749         Elf_Scn *scn = NULL;
 750         int idx = 0, err = 0;
 751
 752         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
 753         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
 754                 pr_warning("failed to get e_shstrndx from %s\n",
 755                            obj->path);
 756                 return -LIBBPF_ERRNO__FORMAT;
 757         }
 758
 759         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 760                 char *name;
 761                 GElf_Shdr sh;
 762                 Elf_Data *data;
 763
 764                 idx++;
 765                 if (gelf_getshdr(scn, &sh) != &sh) {
 766                         pr_warning("failed to get section(%d) header from %s\n",
 767                                    idx, obj->path);
 768                         err = -LIBBPF_ERRNO__FORMAT;
 769                         goto out;
 770                 }
 771
 772                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 773                 if (!name) {
 774                         pr_warning("failed to get section(%d) name from %s\n",
 775                                    idx, obj->path);
 776                         err = -LIBBPF_ERRNO__FORMAT;
 777                         goto out;
 778                 }
 779
 780                 data = elf_getdata(scn, 0);
 781                 if (!data) {
 782                         pr_warning("failed to get section(%d) data from %s(%s)\n",
 783                                    idx, name, obj->path);
 784                         err = -LIBBPF_ERRNO__FORMAT;
 785                         goto out;
 786                 }
 787                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
 788                          idx, name, (unsigned long)data->d_size,
 789                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
 790                          (int)sh.sh_type);
 791
 792                 if (strcmp(name, "license") == 0)
 793                         err = bpf_object__init_license(obj,
 794                                                        data->d_buf,
 795                                                        data->d_size);
 796                 else if (strcmp(name, "version") == 0)
 797                         err = bpf_object__init_kversion(obj,
 798                                                         data->d_buf,
 799                                                         data->d_size);
 800                 else if (strcmp(name, "maps") == 0)
 801                         obj->efile.maps_shndx = idx;
 802                 else if (sh.sh_type == SHT_SYMTAB) {
 803                         if (obj->efile.symbols) {
 804                                 pr_warning("bpf: multiple SYMTAB in %s\n",
 805                                            obj->path);
 806                                 err = -LIBBPF_ERRNO__FORMAT;
 807                         } else {
 808                                 obj->efile.symbols = data;
 809                                 obj->efile.strtabidx = sh.sh_link;
 810                         }
 811                 } else if ((sh.sh_type == SHT_PROGBITS) &&
 812                            (sh.sh_flags & SHF_EXECINSTR) &&
 813                            (data->d_size > 0)) {
 814                         if (strcmp(name, ".text") == 0)
 815                                 obj->efile.text_shndx = idx;
 816                         err = bpf_object__add_program(obj, data->d_buf,
 817                                                       data->d_size, name, idx);
 818                         if (err) {
 819                                 char errmsg[STRERR_BUFSIZE];
 820
 821                                 strerror_r(-err, errmsg, sizeof(errmsg));
 822                                 pr_warning("failed to alloc program %s (%s): %s",
 823                                            name, obj->path, errmsg);
 824                         }
 825                 } else if (sh.sh_type == SHT_REL) {
 826                         void *reloc = obj->efile.reloc;
 827                         int nr_reloc = obj->efile.nr_reloc + 1;
 828
 829                         reloc = realloc(reloc,
 830                                         sizeof(*obj->efile.reloc) * nr_reloc);
 831                         if (!reloc) {
 832                                 pr_warning("realloc failed\n");
 833                                 err = -ENOMEM;
 834                         } else {
 835                                 int n = nr_reloc - 1;
 836
 837                                 obj->efile.reloc = reloc;
 838                                 obj->efile.nr_reloc = nr_reloc;
 839
 840                                 obj->efile.reloc[n].shdr = sh;
 841                                 obj->efile.reloc[n].data = data;
 842                         }
 843                 } else {
 844                         pr_debug("skip section(%d) %s\n", idx, name);
 845                 }
 846                 if (err)
 847                         goto out;
 848         }
 849
 850         if (!obj->efile.strtabidx || obj->efile.strtabidx >= idx) {
 851                 pr_warning("Corrupted ELF file: index of strtab invalid\n");
 852                 return LIBBPF_ERRNO__FORMAT;
 853         }
 854         if (obj->efile.maps_shndx >= 0) {
 855                 err = bpf_object__init_maps(obj);
 856                 if (err)
 857                         goto out;
 858         }
 859         err = bpf_object__init_prog_names(obj);
 860 out:
 861         return err;
 862 }
 863
 864 static struct bpf_program *
 865 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
 866 {
 867         struct bpf_program *prog;
 868         size_t i;
 869
 870         for (i = 0; i < obj->nr_programs; i++) {
 871                 prog = &obj->programs[i];
 872                 if (prog->idx == idx)
 873                         return prog;
 874         }
 875         return NULL;
 876 }
 877
 878 static int
 879 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
 880                            Elf_Data *data, struct bpf_object *obj)
 881 {
 882         Elf_Data *symbols = obj->efile.symbols;
 883         int text_shndx = obj->efile.text_shndx;
 884         int maps_shndx = obj->efile.maps_shndx;
 885         struct bpf_map *maps = obj->maps;
 886         size_t nr_maps = obj->nr_maps;
 887         int i, nrels;
 888
 889         pr_debug("collecting relocating info for: '%s'\n",
 890                  prog->section_name);
 891         nrels = shdr->sh_size / shdr->sh_entsize;
 892
 893         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
 894         if (!prog->reloc_desc) {
 895                 pr_warning("failed to alloc memory in relocation\n");
 896                 return -ENOMEM;
 897         }
 898         prog->nr_reloc = nrels;
 899
 900         for (i = 0; i < nrels; i++) {
 901                 GElf_Sym sym;
 902                 GElf_Rel rel;
 903                 unsigned int insn_idx;
 904                 struct bpf_insn *insns = prog->insns;
 905                 size_t map_idx;
 906
 907                 if (!gelf_getrel(data, i, &rel)) {
 908                         pr_warning("relocation: failed to get %d reloc\n", i);
 909                         return -LIBBPF_ERRNO__FORMAT;
 910                 }
 911
 912                 if (!gelf_getsym(symbols,
 913                                  GELF_R_SYM(rel.r_info),
 914                                  &sym)) {
 915                         pr_warning("relocation: symbol %"PRIx64" not found\n",
 916                                    GELF_R_SYM(rel.r_info));
 917                         return -LIBBPF_ERRNO__FORMAT;
 918                 }
 919                 pr_debug("relo for %lld value %lld name %d\n",
 920                          (long long) (rel.r_info >> 32),
 921                          (long long) sym.st_value, sym.st_name);
 922
 923                 if (sym.st_shndx != maps_shndx && sym.st_shndx != text_shndx) {
 924                         pr_warning("Program '%s' contains non-map related relo data pointing to section %u\n",
 925                                    prog->section_name, sym.st_shndx);
 926                         return -LIBBPF_ERRNO__RELOC;
 927                 }
 928
 929                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
 930                 pr_debug("relocation: insn_idx=%u\n", insn_idx);
 931
 932                 if (insns[insn_idx].code == (BPF_JMP | BPF_CALL)) {
 933                         if (insns[insn_idx].src_reg != BPF_PSEUDO_CALL) {
 934                                 pr_warning("incorrect bpf_call opcode\n");
 935                                 return -LIBBPF_ERRNO__RELOC;
 936                         }
 937                         prog->reloc_desc[i].type = RELO_CALL;
 938                         prog->reloc_desc[i].insn_idx = insn_idx;
 939                         prog->reloc_desc[i].text_off = sym.st_value;
 940                         continue;
 941                 }
 942
 943                 if (insns[insn_idx].code != (BPF_LD | BPF_IMM | BPF_DW)) {
 944                         pr_warning("bpf: relocation: invalid relo for insns[%d].code 0x%x\n",
 945                                    insn_idx, insns[insn_idx].code);
 946                         return -LIBBPF_ERRNO__RELOC;
 947                 }
 948
 949                 /* TODO: 'maps' is sorted. We can use bsearch to make it faster. */
 950                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
 951                         if (maps[map_idx].offset == sym.st_value) {
 952                                 pr_debug("relocation: find map %zd (%s) for insn %u\n",
 953                                          map_idx, maps[map_idx].name, insn_idx);
 954                                 break;
 955                         }
 956                 }
 957
 958                 if (map_idx >= nr_maps) {
 959                         pr_warning("bpf relocation: map_idx %d large than %d\n",
 960                                    (int)map_idx, (int)nr_maps - 1);
 961                         return -LIBBPF_ERRNO__RELOC;
 962                 }
 963
 964                 prog->reloc_desc[i].type = RELO_LD64;
 965                 prog->reloc_desc[i].insn_idx = insn_idx;
 966                 prog->reloc_desc[i].map_idx = map_idx;
 967         }
 968         return 0;
 969 }
 970
 971 static int
 972 bpf_object__create_maps(struct bpf_object *obj)
 973 {
 974         unsigned int i;
 975
 976         for (i = 0; i < obj->nr_maps; i++) {
 977                 struct bpf_map_def *def = &obj->maps[i].def;
 978                 int *pfd = &obj->maps[i].fd;
 979
 980                 *pfd = bpf_create_map_name(def->type,
 981                                            obj->maps[i].name,
 982                                            def->key_size,
 983                                            def->value_size,
 984                                            def->max_entries,
 985                                            def->map_flags);
 986                 if (*pfd < 0) {
 987                         size_t j;
 988                         int err = *pfd;
 989
 990                         pr_warning("failed to create map (name: '%s'): %s\n",
 991                                    obj->maps[i].name,
 992                                    strerror(errno));
 993                         for (j = 0; j < i; j++)
 994                                 zclose(obj->maps[j].fd);
 995                         return err;
 996                 }
 997                 pr_debug("create map %s: fd=%d\n", obj->maps[i].name, *pfd);
 998         }
 999
1000         return 0;
1001 }
1002
1003 static int
1004 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
1005                         struct reloc_desc *relo)
1006 {
1007         struct bpf_insn *insn, *new_insn;
1008         struct bpf_program *text;
1009         size_t new_cnt;
1010
1011         if (relo->type != RELO_CALL)
1012                 return -LIBBPF_ERRNO__RELOC;
1013
1014         if (prog->idx == obj->efile.text_shndx) {
1015                 pr_warning("relo in .text insn %d into off %d\n",
1016                            relo->insn_idx, relo->text_off);
1017                 return -LIBBPF_ERRNO__RELOC;
1018         }
1019
1020         if (prog->main_prog_cnt == 0) {
1021                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
1022                 if (!text) {
1023                         pr_warning("no .text section found yet relo into text exist\n");
1024                         return -LIBBPF_ERRNO__RELOC;
1025                 }
1026                 new_cnt = prog->insns_cnt + text->insns_cnt;
1027                 new_insn = realloc(prog->insns, new_cnt * sizeof(*insn));
1028                 if (!new_insn) {
1029                         pr_warning("oom in prog realloc\n");
1030                         return -ENOMEM;
1031                 }
1032                 memcpy(new_insn + prog->insns_cnt, text->insns,
1033                        text->insns_cnt * sizeof(*insn));
1034                 prog->insns = new_insn;
1035                 prog->main_prog_cnt = prog->insns_cnt;
1036                 prog->insns_cnt = new_cnt;
1037         }
1038         insn = &prog->insns[relo->insn_idx];
1039         insn->imm += prog->main_prog_cnt - relo->insn_idx;
1040         pr_debug("added %zd insn from %s to prog %s\n",
1041                  text->insns_cnt, text->section_name, prog->section_name);
1042         return 0;
1043 }
1044
1045 static int
1046 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
1047 {
1048         int i, err;
1049
1050         if (!prog || !prog->reloc_desc)
1051                 return 0;
1052
1053         for (i = 0; i < prog->nr_reloc; i++) {
1054                 if (prog->reloc_desc[i].type == RELO_LD64) {
1055                         struct bpf_insn *insns = prog->insns;
1056                         int insn_idx, map_idx;
1057
1058                         insn_idx = prog->reloc_desc[i].insn_idx;
1059                         map_idx = prog->reloc_desc[i].map_idx;
1060
1061                         if (insn_idx >= (int)prog->insns_cnt) {
1062                                 pr_warning("relocation out of range: '%s'\n",
1063                                            prog->section_name);
1064                                 return -LIBBPF_ERRNO__RELOC;
1065                         }
1066                         insns[insn_idx].src_reg = BPF_PSEUDO_MAP_FD;
1067                         insns[insn_idx].imm = obj->maps[map_idx].fd;
1068                 } else {
1069                         err = bpf_program__reloc_text(prog, obj,
1070                                                       &prog->reloc_desc[i]);
1071                         if (err)
1072                                 return err;
1073                 }
1074         }
1075
1076         zfree(&prog->reloc_desc);
1077         prog->nr_reloc = 0;
1078         return 0;
1079 }
1080
1081
1082 static int
1083 bpf_object__relocate(struct bpf_object *obj)
1084 {
1085         struct bpf_program *prog;
1086         size_t i;
1087         int err;
1088
1089         for (i = 0; i < obj->nr_programs; i++) {
1090                 prog = &obj->programs[i];
1091
1092                 err = bpf_program__relocate(prog, obj);
1093                 if (err) {
1094                         pr_warning("failed to relocate '%s'\n",
1095                                    prog->section_name);
1096                         return err;
1097                 }
1098         }
1099         return 0;
1100 }
1101
1102 static int bpf_object__collect_reloc(struct bpf_object *obj)
1103 {
1104         int i, err;
1105
1106         if (!obj_elf_valid(obj)) {
1107                 pr_warning("Internal error: elf object is closed\n");
1108                 return -LIBBPF_ERRNO__INTERNAL;
1109         }
1110
1111         for (i = 0; i < obj->efile.nr_reloc; i++) {
1112                 GElf_Shdr *shdr = &obj->efile.reloc[i].shdr;
1113                 Elf_Data *data = obj->efile.reloc[i].data;
1114                 int idx = shdr->sh_info;
1115                 struct bpf_program *prog;
1116
1117                 if (shdr->sh_type != SHT_REL) {
1118                         pr_warning("internal error at %d\n", __LINE__);
1119                         return -LIBBPF_ERRNO__INTERNAL;
1120                 }
1121
1122                 prog = bpf_object__find_prog_by_idx(obj, idx);
1123                 if (!prog) {
1124                         pr_warning("relocation failed: no section(%d)\n", idx);
1125                         return -LIBBPF_ERRNO__RELOC;
1126                 }
1127
1128                 err = bpf_program__collect_reloc(prog,
1129                                                  shdr, data,
1130                                                  obj);
1131                 if (err)
1132                         return err;
1133         }
1134         return 0;
1135 }
1136
1137 static int
1138 load_program(enum bpf_prog_type type, const char *name, struct bpf_insn *insns,
1139              int insns_cnt, char *license, u32 kern_version, int *pfd)
1140 {
1141         int ret;
1142         char *log_buf;
1143
1144         if (!insns || !insns_cnt)
1145                 return -EINVAL;
1146
1147         log_buf = malloc(BPF_LOG_BUF_SIZE);
1148         if (!log_buf)
1149                 pr_warning("Alloc log buffer for bpf loader error, continue without log\n");
1150
1151         ret = bpf_load_program_name(type, name, insns, insns_cnt, license,
1152                                     kern_version, log_buf, BPF_LOG_BUF_SIZE);
1153
1154         if (ret >= 0) {
1155                 *pfd = ret;
1156                 ret = 0;
1157                 goto out;
1158         }
1159
1160         ret = -LIBBPF_ERRNO__LOAD;
1161         pr_warning("load bpf program failed: %s\n", strerror(errno));
1162
1163         if (log_buf && log_buf[0] != '\0') {
1164                 ret = -LIBBPF_ERRNO__VERIFY;
1165                 pr_warning("-- BEGIN DUMP LOG ---\n");
1166                 pr_warning("\n%s\n", log_buf);
1167                 pr_warning("-- END LOG --\n");
1168         } else if (insns_cnt >= BPF_MAXINSNS) {
1169                 pr_warning("Program too large (%d insns), at most %d insns\n",
1170                            insns_cnt, BPF_MAXINSNS);
1171                 ret = -LIBBPF_ERRNO__PROG2BIG;
1172         } else {
1173                 /* Wrong program type? */
1174                 if (type != BPF_PROG_TYPE_KPROBE) {
1175                         int fd;
1176
1177                         fd = bpf_load_program_name(BPF_PROG_TYPE_KPROBE, name,
1178                                                    insns, insns_cnt, license,
1179                                                    kern_version, NULL, 0);
1180                         if (fd >= 0) {
1181                                 close(fd);
1182                                 ret = -LIBBPF_ERRNO__PROGTYPE;
1183                                 goto out;
1184                         }
1185                 }
1186
1187                 if (log_buf)
1188                         ret = -LIBBPF_ERRNO__KVER;
1189         }
1190
1191 out:
1192         free(log_buf);
1193         return ret;
1194 }
1195
1196 static int
1197 bpf_program__load(struct bpf_program *prog,
1198                   char *license, u32 kern_version)
1199 {
1200         int err = 0, fd, i;
1201
1202         if (prog->instances.nr < 0 || !prog->instances.fds) {
1203                 if (prog->preprocessor) {
1204                         pr_warning("Internal error: can't load program '%s'\n",
1205                                    prog->section_name);
1206                         return -LIBBPF_ERRNO__INTERNAL;
1207                 }
1208
1209                 prog->instances.fds = malloc(sizeof(int));
1210                 if (!prog->instances.fds) {
1211                         pr_warning("Not enough memory for BPF fds\n");
1212                         return -ENOMEM;
1213                 }
1214                 prog->instances.nr = 1;
1215                 prog->instances.fds[0] = -1;
1216         }
1217
1218         if (!prog->preprocessor) {
1219                 if (prog->instances.nr != 1) {
1220                         pr_warning("Program '%s' is inconsistent: nr(%d) != 1\n",
1221                                    prog->section_name, prog->instances.nr);
1222                 }
1223                 err = load_program(prog->type, prog->name, prog->insns,
1224                                    prog->insns_cnt, license, kern_version, &fd);
1225                 if (!err)
1226                         prog->instances.fds[0] = fd;
1227                 goto out;
1228         }
1229
1230         for (i = 0; i < prog->instances.nr; i++) {
1231                 struct bpf_prog_prep_result result;
1232                 bpf_program_prep_t preprocessor = prog->preprocessor;
1233
1234                 bzero(&result, sizeof(result));
1235                 err = preprocessor(prog, i, prog->insns,
1236                                    prog->insns_cnt, &result);
1237                 if (err) {
1238                         pr_warning("Preprocessing the %dth instance of program '%s' failed\n",
1239                                    i, prog->section_name);
1240                         goto out;
1241                 }
1242
1243                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
1244                         pr_debug("Skip loading the %dth instance of program '%s'\n",
1245                                  i, prog->section_name);
1246                         prog->instances.fds[i] = -1;
1247                         if (result.pfd)
1248                                 *result.pfd = -1;
1249                         continue;
1250                 }
1251
1252                 err = load_program(prog->type, prog->name,
1253                                    result.new_insn_ptr,
1254                                    result.new_insn_cnt,
1255                                    license, kern_version, &fd);
1256
1257                 if (err) {
1258                         pr_warning("Loading the %dth instance of program '%s' failed\n",
1259                                         i, prog->section_name);
1260                         goto out;
1261                 }
1262
1263                 if (result.pfd)
1264                         *result.pfd = fd;
1265                 prog->instances.fds[i] = fd;
1266         }
1267 out:
1268         if (err)
1269                 pr_warning("failed to load program '%s'\n",
1270                            prog->section_name);
1271         zfree(&prog->insns);
1272         prog->insns_cnt = 0;
1273         return err;
1274 }
1275
1276 static int
1277 bpf_object__load_progs(struct bpf_object *obj)
1278 {
1279         size_t i;
1280         int err;
1281
1282         for (i = 0; i < obj->nr_programs; i++) {
1283                 if (obj->programs[i].idx == obj->efile.text_shndx)
1284                         continue;
1285                 err = bpf_program__load(&obj->programs[i],
1286                                         obj->license,
1287                                         obj->kern_version);
1288                 if (err)
1289                         return err;
1290         }
1291         return 0;
1292 }
1293
1294 static int bpf_object__validate(struct bpf_object *obj)
1295 {
1296         if (obj->kern_version == 0) {
1297                 pr_warning("%s doesn't provide kernel version\n",
1298                            obj->path);
1299                 return -LIBBPF_ERRNO__KVERSION;
1300         }
1301         return 0;
1302 }
1303
1304 static struct bpf_object *
1305 __bpf_object__open(const char *path, void *obj_buf, size_t obj_buf_sz)
1306 {
1307         struct bpf_object *obj;
1308         int err;
1309
1310         if (elf_version(EV_CURRENT) == EV_NONE) {
1311                 pr_warning("failed to init libelf for %s\n", path);
1312                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1313         }
1314
1315         obj = bpf_object__new(path, obj_buf, obj_buf_sz);
1316         if (IS_ERR(obj))
1317                 return obj;
1318
1319         CHECK_ERR(bpf_object__elf_init(obj), err, out);
1320         CHECK_ERR(bpf_object__check_endianness(obj), err, out);
1321         CHECK_ERR(bpf_object__elf_collect(obj), err, out);
1322         CHECK_ERR(bpf_object__collect_reloc(obj), err, out);
1323         CHECK_ERR(bpf_object__validate(obj), err, out);
1324
1325         bpf_object__elf_finish(obj);
1326         return obj;
1327 out:
1328         bpf_object__close(obj);
1329         return ERR_PTR(err);
1330 }
1331
1332 struct bpf_object *bpf_object__open(const char *path)
1333 {
1334         /* param validation */
1335         if (!path)
1336                 return NULL;
1337
1338         pr_debug("loading %s\n", path);
1339
1340         return __bpf_object__open(path, NULL, 0);
1341 }
1342
1343 struct bpf_object *bpf_object__open_buffer(void *obj_buf,
1344                                            size_t obj_buf_sz,
1345                                            const char *name)
1346 {
1347         char tmp_name[64];
1348
1349         /* param validation */
1350         if (!obj_buf || obj_buf_sz <= 0)
1351                 return NULL;
1352
1353         if (!name) {
1354                 snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
1355                          (unsigned long)obj_buf,
1356                          (unsigned long)obj_buf_sz);
1357                 tmp_name[sizeof(tmp_name) - 1] = '\0';
1358                 name = tmp_name;
1359         }
1360         pr_debug("loading object '%s' from buffer\n",
1361                  name);
1362
1363         return __bpf_object__open(name, obj_buf, obj_buf_sz);
1364 }
1365
1366 int bpf_object__unload(struct bpf_object *obj)
1367 {
1368         size_t i;
1369
1370         if (!obj)
1371                 return -EINVAL;
1372
1373         for (i = 0; i < obj->nr_maps; i++)
1374                 zclose(obj->maps[i].fd);
1375
1376         for (i = 0; i < obj->nr_programs; i++)
1377                 bpf_program__unload(&obj->programs[i]);
1378
1379         return 0;
1380 }
1381
1382 int bpf_object__load(struct bpf_object *obj)
1383 {
1384         int err;
1385
1386         if (!obj)
1387                 return -EINVAL;
1388
1389         if (obj->loaded) {
1390                 pr_warning("object should not be loaded twice\n");
1391                 return -EINVAL;
1392         }
1393
1394         obj->loaded = true;
1395
1396         CHECK_ERR(bpf_object__create_maps(obj), err, out);
1397         CHECK_ERR(bpf_object__relocate(obj), err, out);
1398         CHECK_ERR(bpf_object__load_progs(obj), err, out);
1399
1400         return 0;
1401 out:
1402         bpf_object__unload(obj);
1403         pr_warning("failed to load object '%s'\n", obj->path);
1404         return err;
1405 }
1406
1407 static int check_path(const char *path)
1408 {
1409         struct statfs st_fs;
1410         char *dname, *dir;
1411         int err = 0;
1412
1413         if (path == NULL)
1414                 return -EINVAL;
1415
1416         dname = strdup(path);
1417         if (dname == NULL)
1418                 return -ENOMEM;
1419
1420         dir = dirname(dname);
1421         if (statfs(dir, &st_fs)) {
1422                 pr_warning("failed to statfs %s: %s\n", dir, strerror(errno));
1423                 err = -errno;
1424         }
1425         free(dname);
1426
1427         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
1428                 pr_warning("specified path %s is not on BPF FS\n", path);
1429                 err = -EINVAL;
1430         }
1431
1432         return err;
1433 }
1434
1435 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
1436                               int instance)
1437 {
1438         int err;
1439
1440         err = check_path(path);
1441         if (err)
1442                 return err;
1443
1444         if (prog == NULL) {
1445                 pr_warning("invalid program pointer\n");
1446                 return -EINVAL;
1447         }
1448
1449         if (instance < 0 || instance >= prog->instances.nr) {
1450                 pr_warning("invalid prog instance %d of prog %s (max %d)\n",
1451                            instance, prog->section_name, prog->instances.nr);
1452                 return -EINVAL;
1453         }
1454
1455         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
1456                 pr_warning("failed to pin program: %s\n", strerror(errno));
1457                 return -errno;
1458         }
1459         pr_debug("pinned program '%s'\n", path);
1460
1461         return 0;
1462 }
1463
1464 static int make_dir(const char *path)
1465 {
1466         int err = 0;
1467
1468         if (mkdir(path, 0700) && errno != EEXIST)
1469                 err = -errno;
1470
1471         if (err)
1472                 pr_warning("failed to mkdir %s: %s\n", path, strerror(-err));
1473         return err;
1474 }
1475
1476 int bpf_program__pin(struct bpf_program *prog, const char *path)
1477 {
1478         int i, err;
1479
1480         err = check_path(path);
1481         if (err)
1482                 return err;
1483
1484         if (prog == NULL) {
1485                 pr_warning("invalid program pointer\n");
1486                 return -EINVAL;
1487         }
1488
1489         if (prog->instances.nr <= 0) {
1490                 pr_warning("no instances of prog %s to pin\n",
1491                            prog->section_name);
1492                 return -EINVAL;
1493         }
1494
1495         err = make_dir(path);
1496         if (err)
1497                 return err;
1498
1499         for (i = 0; i < prog->instances.nr; i++) {
1500                 char buf[PATH_MAX];
1501                 int len;
1502
1503                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
1504                 if (len < 0)
1505                         return -EINVAL;
1506                 else if (len >= PATH_MAX)
1507                         return -ENAMETOOLONG;
1508
1509                 err = bpf_program__pin_instance(prog, buf, i);
1510                 if (err)
1511                         return err;
1512         }
1513
1514         return 0;
1515 }
1516
1517 int bpf_map__pin(struct bpf_map *map, const char *path)
1518 {
1519         int err;
1520
1521         err = check_path(path);
1522         if (err)
1523                 return err;
1524
1525         if (map == NULL) {
1526                 pr_warning("invalid map pointer\n");
1527                 return -EINVAL;
1528         }
1529
1530         if (bpf_obj_pin(map->fd, path)) {
1531                 pr_warning("failed to pin map: %s\n", strerror(errno));
1532                 return -errno;
1533         }
1534
1535         pr_debug("pinned map '%s'\n", path);
1536         return 0;
1537 }
1538
1539 int bpf_object__pin(struct bpf_object *obj, const char *path)
1540 {
1541         struct bpf_program *prog;
1542         struct bpf_map *map;
1543         int err;
1544
1545         if (!obj)
1546                 return -ENOENT;
1547
1548         if (!obj->loaded) {
1549                 pr_warning("object not yet loaded; load it first\n");
1550                 return -ENOENT;
1551         }
1552
1553         err = make_dir(path);
1554         if (err)
1555                 return err;
1556
1557         bpf_map__for_each(map, obj) {
1558                 char buf[PATH_MAX];
1559                 int len;
1560
1561                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1562                                bpf_map__name(map));
1563                 if (len < 0)
1564                         return -EINVAL;
1565                 else if (len >= PATH_MAX)
1566                         return -ENAMETOOLONG;
1567
1568                 err = bpf_map__pin(map, buf);
1569                 if (err)
1570                         return err;
1571         }
1572
1573         bpf_object__for_each_program(prog, obj) {
1574                 char buf[PATH_MAX];
1575                 int len;
1576
1577                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
1578                                prog->section_name);
1579                 if (len < 0)
1580                         return -EINVAL;
1581                 else if (len >= PATH_MAX)
1582                         return -ENAMETOOLONG;
1583
1584                 err = bpf_program__pin(prog, buf);
1585                 if (err)
1586                         return err;
1587         }
1588
1589         return 0;
1590 }
1591
1592 void bpf_object__close(struct bpf_object *obj)
1593 {
1594         size_t i;
1595
1596         if (!obj)
1597                 return;
1598
1599         if (obj->clear_priv)
1600                 obj->clear_priv(obj, obj->priv);
1601
1602         bpf_object__elf_finish(obj);
1603         bpf_object__unload(obj);
1604
1605         for (i = 0; i < obj->nr_maps; i++) {
1606                 zfree(&obj->maps[i].name);
1607                 if (obj->maps[i].clear_priv)
1608                         obj->maps[i].clear_priv(&obj->maps[i],
1609                                                 obj->maps[i].priv);
1610                 obj->maps[i].priv = NULL;
1611                 obj->maps[i].clear_priv = NULL;
1612         }
1613         zfree(&obj->maps);
1614         obj->nr_maps = 0;
1615
1616         if (obj->programs && obj->nr_programs) {
1617                 for (i = 0; i < obj->nr_programs; i++)
1618                         bpf_program__exit(&obj->programs[i]);
1619         }
1620         zfree(&obj->programs);
1621
1622         list_del(&obj->list);
1623         free(obj);
1624 }
1625
1626 struct bpf_object *
1627 bpf_object__next(struct bpf_object *prev)
1628 {
1629         struct bpf_object *next;
1630
1631         if (!prev)
1632                 next = list_first_entry(&bpf_objects_list,
1633                                         struct bpf_object,
1634                                         list);
1635         else
1636                 next = list_next_entry(prev, list);
1637
1638         /* Empty list is noticed here so don't need checking on entry. */
1639         if (&next->list == &bpf_objects_list)
1640                 return NULL;
1641
1642         return next;
1643 }
1644
1645 const char *bpf_object__name(struct bpf_object *obj)
1646 {
1647         return obj ? obj->path : ERR_PTR(-EINVAL);
1648 }
1649
1650 unsigned int bpf_object__kversion(struct bpf_object *obj)
1651 {
1652         return obj ? obj->kern_version : 0;
1653 }
1654
1655 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
1656                          bpf_object_clear_priv_t clear_priv)
1657 {
1658         if (obj->priv && obj->clear_priv)
1659                 obj->clear_priv(obj, obj->priv);
1660
1661         obj->priv = priv;
1662         obj->clear_priv = clear_priv;
1663         return 0;
1664 }
1665
1666 void *bpf_object__priv(struct bpf_object *obj)
1667 {
1668         return obj ? obj->priv : ERR_PTR(-EINVAL);
1669 }
1670
1671 struct bpf_program *
1672 bpf_program__next(struct bpf_program *prev, struct bpf_object *obj)
1673 {
1674         size_t idx;
1675
1676         if (!obj->programs)
1677                 return NULL;
1678         /* First handler */
1679         if (prev == NULL)
1680                 return &obj->programs[0];
1681
1682         if (prev->obj != obj) {
1683                 pr_warning("error: program handler doesn't match object\n");
1684                 return NULL;
1685         }
1686
1687         idx = (prev - obj->programs) + 1;
1688         if (idx >= obj->nr_programs)
1689                 return NULL;
1690         return &obj->programs[idx];
1691 }
1692
1693 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
1694                           bpf_program_clear_priv_t clear_priv)
1695 {
1696         if (prog->priv && prog->clear_priv)
1697                 prog->clear_priv(prog, prog->priv);
1698
1699         prog->priv = priv;
1700         prog->clear_priv = clear_priv;
1701         return 0;
1702 }
1703
1704 void *bpf_program__priv(struct bpf_program *prog)
1705 {
1706         return prog ? prog->priv : ERR_PTR(-EINVAL);
1707 }
1708
1709 const char *bpf_program__title(struct bpf_program *prog, bool needs_copy)
1710 {
1711         const char *title;
1712
1713         title = prog->section_name;
1714         if (needs_copy) {
1715                 title = strdup(title);
1716                 if (!title) {
1717                         pr_warning("failed to strdup program title\n");
1718                         return ERR_PTR(-ENOMEM);
1719                 }
1720         }
1721
1722         return title;
1723 }
1724
1725 int bpf_program__fd(struct bpf_program *prog)
1726 {
1727         return bpf_program__nth_fd(prog, 0);
1728 }
1729
1730 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
1731                           bpf_program_prep_t prep)
1732 {
1733         int *instances_fds;
1734
1735         if (nr_instances <= 0 || !prep)
1736                 return -EINVAL;
1737
1738         if (prog->instances.nr > 0 || prog->instances.fds) {
1739                 pr_warning("Can't set pre-processor after loading\n");
1740                 return -EINVAL;
1741         }
1742
1743         instances_fds = malloc(sizeof(int) * nr_instances);
1744         if (!instances_fds) {
1745                 pr_warning("alloc memory failed for fds\n");
1746                 return -ENOMEM;
1747         }
1748
1749         /* fill all fd with -1 */
1750         memset(instances_fds, -1, sizeof(int) * nr_instances);
1751
1752         prog->instances.nr = nr_instances;
1753         prog->instances.fds = instances_fds;
1754         prog->preprocessor = prep;
1755         return 0;
1756 }
1757
1758 int bpf_program__nth_fd(struct bpf_program *prog, int n)
1759 {
1760         int fd;
1761
1762         if (n >= prog->instances.nr || n < 0) {
1763                 pr_warning("Can't get the %dth fd from program %s: only %d instances\n",
1764                            n, prog->section_name, prog->instances.nr);
1765                 return -EINVAL;
1766         }
1767
1768         fd = prog->instances.fds[n];
1769         if (fd < 0) {
1770                 pr_warning("%dth instance of program '%s' is invalid\n",
1771                            n, prog->section_name);
1772                 return -ENOENT;
1773         }
1774
1775         return fd;
1776 }
1777
1778 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
1779 {
1780         prog->type = type;
1781 }
1782
1783 static bool bpf_program__is_type(struct bpf_program *prog,
1784                                  enum bpf_prog_type type)
1785 {
1786         return prog ? (prog->type == type) : false;
1787 }
1788
1789 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                   \
1790 int bpf_program__set_##NAME(struct bpf_program *prog)   \
1791 {                                                       \
1792         if (!prog)                                      \
1793                 return -EINVAL;                         \
1794         bpf_program__set_type(prog, TYPE);              \
1795         return 0;                                       \
1796 }                                                       \
1797                                                         \
1798 bool bpf_program__is_##NAME(struct bpf_program *prog)   \
1799 {                                                       \
1800         return bpf_program__is_type(prog, TYPE);        \
1801 }                                                       \
1802
1803 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
1804 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
1805 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
1806 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
1807 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
1808 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
1809 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
1810
1811 #define BPF_PROG_SEC(string, type) { string, sizeof(string) - 1, type }
1812 static const struct {
1813         const char *sec;
1814         size_t len;
1815         enum bpf_prog_type prog_type;
1816 } section_names[] = {
1817         BPF_PROG_SEC("socket",          BPF_PROG_TYPE_SOCKET_FILTER),
1818         BPF_PROG_SEC("kprobe/",         BPF_PROG_TYPE_KPROBE),
1819         BPF_PROG_SEC("kretprobe/",      BPF_PROG_TYPE_KPROBE),
1820         BPF_PROG_SEC("classifier",      BPF_PROG_TYPE_SCHED_CLS),
1821         BPF_PROG_SEC("action",          BPF_PROG_TYPE_SCHED_ACT),
1822         BPF_PROG_SEC("tracepoint/",     BPF_PROG_TYPE_TRACEPOINT),
1823         BPF_PROG_SEC("xdp",             BPF_PROG_TYPE_XDP),
1824         BPF_PROG_SEC("perf_event",      BPF_PROG_TYPE_PERF_EVENT),
1825         BPF_PROG_SEC("cgroup/skb",      BPF_PROG_TYPE_CGROUP_SKB),
1826         BPF_PROG_SEC("cgroup/sock",     BPF_PROG_TYPE_CGROUP_SOCK),
1827         BPF_PROG_SEC("cgroup/dev",      BPF_PROG_TYPE_CGROUP_DEVICE),
1828         BPF_PROG_SEC("lwt_in",          BPF_PROG_TYPE_LWT_IN),
1829         BPF_PROG_SEC("lwt_out",         BPF_PROG_TYPE_LWT_OUT),
1830         BPF_PROG_SEC("lwt_xmit",        BPF_PROG_TYPE_LWT_XMIT),
1831         BPF_PROG_SEC("sockops",         BPF_PROG_TYPE_SOCK_OPS),
1832         BPF_PROG_SEC("sk_skb",          BPF_PROG_TYPE_SK_SKB),
1833 };
1834 #undef BPF_PROG_SEC
1835
1836 static enum bpf_prog_type bpf_program__guess_type(struct bpf_program *prog)
1837 {
1838         int i;
1839
1840         if (!prog->section_name)
1841                 goto err;
1842
1843         for (i = 0; i < ARRAY_SIZE(section_names); i++)
1844                 if (strncmp(prog->section_name, section_names[i].sec,
1845                             section_names[i].len) == 0)
1846                         return section_names[i].prog_type;
1847
1848 err:
1849         pr_warning("failed to guess program type based on section name %s\n",
1850                    prog->section_name);
1851
1852         return BPF_PROG_TYPE_UNSPEC;
1853 }
1854
1855 int bpf_map__fd(struct bpf_map *map)
1856 {
1857         return map ? map->fd : -EINVAL;
1858 }
1859
1860 const struct bpf_map_def *bpf_map__def(struct bpf_map *map)
1861 {
1862         return map ? &map->def : ERR_PTR(-EINVAL);
1863 }
1864
1865 const char *bpf_map__name(struct bpf_map *map)
1866 {
1867         return map ? map->name : NULL;
1868 }
1869
1870 int bpf_map__set_priv(struct bpf_map *map, void *priv,
1871                      bpf_map_clear_priv_t clear_priv)
1872 {
1873         if (!map)
1874                 return -EINVAL;
1875
1876         if (map->priv) {
1877                 if (map->clear_priv)
1878                         map->clear_priv(map, map->priv);
1879         }
1880
1881         map->priv = priv;
1882         map->clear_priv = clear_priv;
1883         return 0;
1884 }
1885
1886 void *bpf_map__priv(struct bpf_map *map)
1887 {
1888         return map ? map->priv : ERR_PTR(-EINVAL);
1889 }
1890
1891 struct bpf_map *
1892 bpf_map__next(struct bpf_map *prev, struct bpf_object *obj)
1893 {
1894         size_t idx;
1895         struct bpf_map *s, *e;
1896
1897         if (!obj || !obj->maps)
1898                 return NULL;
1899
1900         s = obj->maps;
1901         e = obj->maps + obj->nr_maps;
1902
1903         if (prev == NULL)
1904                 return s;
1905
1906         if ((prev < s) || (prev >= e)) {
1907                 pr_warning("error in %s: map handler doesn't belong to object\n",
1908                            __func__);
1909                 return NULL;
1910         }
1911
1912         idx = (prev - obj->maps) + 1;
1913         if (idx >= obj->nr_maps)
1914                 return NULL;
1915         return &obj->maps[idx];
1916 }
1917
1918 struct bpf_map *
1919 bpf_object__find_map_by_name(struct bpf_object *obj, const char *name)
1920 {
1921         struct bpf_map *pos;
1922
1923         bpf_map__for_each(pos, obj) {
1924                 if (pos->name && !strcmp(pos->name, name))
1925                         return pos;
1926         }
1927         return NULL;
1928 }
1929
1930 struct bpf_map *
1931 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
1932 {
1933         int i;
1934
1935         for (i = 0; i < obj->nr_maps; i++) {
1936                 if (obj->maps[i].offset == offset)
1937                         return &obj->maps[i];
1938         }
1939         return ERR_PTR(-ENOENT);
1940 }
1941
1942 long libbpf_get_error(const void *ptr)
1943 {
1944         if (IS_ERR(ptr))
1945                 return PTR_ERR(ptr);
1946         return 0;
1947 }
1948
1949 int bpf_prog_load(const char *file, enum bpf_prog_type type,
1950                   struct bpf_object **pobj, int *prog_fd)
1951 {
1952         struct bpf_program *prog, *first_prog = NULL;
1953         struct bpf_object *obj;
1954         int err;
1955
1956         obj = bpf_object__open(file);
1957         if (IS_ERR(obj))
1958                 return -ENOENT;
1959
1960         bpf_object__for_each_program(prog, obj) {
1961                 /*
1962                  * If type is not specified, try to guess it based on
1963                  * section name.
1964                  */
1965                 if (type == BPF_PROG_TYPE_UNSPEC) {
1966                         type = bpf_program__guess_type(prog);
1967                         if (type == BPF_PROG_TYPE_UNSPEC) {
1968                                 bpf_object__close(obj);
1969                                 return -EINVAL;
1970                         }
1971                 }
1972
1973                 bpf_program__set_type(prog, type);
1974                 if (prog->idx != obj->efile.text_shndx && !first_prog)
1975                         first_prog = prog;
1976         }
1977
1978         if (!first_prog) {
1979                 pr_warning("object file doesn't contain bpf program\n");
1980                 bpf_object__close(obj);
1981                 return -ENOENT;
1982         }
1983
1984         err = bpf_object__load(obj);
1985         if (err) {
1986                 bpf_object__close(obj);
1987                 return -EINVAL;
1988         }
1989
1990         *pobj = obj;
1991         *prog_fd = bpf_program__fd(first_prog);
1992         return 0;
1993 }