tools/lib/bpf/libbpf.c

   1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
   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  * Copyright (C) 2019 Isovalent, Inc.
  11  */
  12
  13 #ifndef _GNU_SOURCE
  14 #define _GNU_SOURCE
  15 #endif
  16 #include <stdlib.h>
  17 #include <stdio.h>
  18 #include <stdarg.h>
  19 #include <libgen.h>
  20 #include <inttypes.h>
  21 #include <limits.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <endian.h>
  25 #include <fcntl.h>
  26 #include <errno.h>
  27 #include <asm/unistd.h>
  28 #include <linux/err.h>
  29 #include <linux/kernel.h>
  30 #include <linux/bpf.h>
  31 #include <linux/btf.h>
  32 #include <linux/filter.h>
  33 #include <linux/list.h>
  34 #include <linux/limits.h>
  35 #include <linux/perf_event.h>
  36 #include <linux/ring_buffer.h>
  37 #include <linux/version.h>
  38 #include <sys/epoll.h>
  39 #include <sys/ioctl.h>
  40 #include <sys/mman.h>
  41 #include <sys/stat.h>
  42 #include <sys/types.h>
  43 #include <sys/vfs.h>
  44 #include <sys/utsname.h>
  45 #include <sys/resource.h>
  46 #include <tools/libc_compat.h>
  47 #include <libelf.h>
  48 #include <gelf.h>
  49 #include <zlib.h>
  50
  51 #include "libbpf.h"
  52 #include "bpf.h"
  53 #include "btf.h"
  54 #include "str_error.h"
  55 #include "libbpf_internal.h"
  56 #include "hashmap.h"
  57
  58 #ifndef EM_BPF
  59 #define EM_BPF 247
  60 #endif
  61
  62 #ifndef BPF_FS_MAGIC
  63 #define BPF_FS_MAGIC            0xcafe4a11
  64 #endif
  65
  66 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
  67  * compilation if user enables corresponding warning. Disable it explicitly.
  68  */
  69 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
  70
  71 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  72
  73 static int __base_pr(enum libbpf_print_level level, const char *format,
  74                      va_list args)
  75 {
  76         if (level == LIBBPF_DEBUG)
  77                 return 0;
  78
  79         return vfprintf(stderr, format, args);
  80 }
  81
  82 static libbpf_print_fn_t __libbpf_pr = __base_pr;
  83
  84 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
  85 {
  86         libbpf_print_fn_t old_print_fn = __libbpf_pr;
  87
  88         __libbpf_pr = fn;
  89         return old_print_fn;
  90 }
  91
  92 __printf(2, 3)
  93 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
  94 {
  95         va_list args;
  96
  97         if (!__libbpf_pr)
  98                 return;
  99
 100         va_start(args, format);
 101         __libbpf_pr(level, format, args);
 102         va_end(args);
 103 }
 104
 105 static void pr_perm_msg(int err)
 106 {
 107         struct rlimit limit;
 108         char buf[100];
 109
 110         if (err != -EPERM || geteuid() != 0)
 111                 return;
 112
 113         err = getrlimit(RLIMIT_MEMLOCK, &limit);
 114         if (err)
 115                 return;
 116
 117         if (limit.rlim_cur == RLIM_INFINITY)
 118                 return;
 119
 120         if (limit.rlim_cur < 1024)
 121                 snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur);
 122         else if (limit.rlim_cur < 1024*1024)
 123                 snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024);
 124         else
 125                 snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024));
 126
 127         pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n",
 128                 buf);
 129 }
 130
 131 #define STRERR_BUFSIZE  128
 132
 133 /* Copied from tools/perf/util/util.h */
 134 #ifndef zfree
 135 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 136 #endif
 137
 138 #ifndef zclose
 139 # define zclose(fd) ({                  \
 140         int ___err = 0;                 \
 141         if ((fd) >= 0)                  \
 142                 ___err = close((fd));   \
 143         fd = -1;                        \
 144         ___err; })
 145 #endif
 146
 147 #ifdef HAVE_LIBELF_MMAP_SUPPORT
 148 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ_MMAP
 149 #else
 150 # define LIBBPF_ELF_C_READ_MMAP ELF_C_READ
 151 #endif
 152
 153 static inline __u64 ptr_to_u64(const void *ptr)
 154 {
 155         return (__u64) (unsigned long) ptr;
 156 }
 157
 158 struct bpf_capabilities {
 159         /* v4.14: kernel support for program & map names. */
 160         __u32 name:1;
 161         /* v5.2: kernel support for global data sections. */
 162         __u32 global_data:1;
 163         /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
 164         __u32 btf_func:1;
 165         /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
 166         __u32 btf_datasec:1;
 167         /* BPF_F_MMAPABLE is supported for arrays */
 168         __u32 array_mmap:1;
 169 };
 170
 171 enum reloc_type {
 172         RELO_LD64,
 173         RELO_CALL,
 174         RELO_DATA,
 175         RELO_EXTERN,
 176 };
 177
 178 struct reloc_desc {
 179         enum reloc_type type;
 180         int insn_idx;
 181         int map_idx;
 182         int sym_off;
 183 };
 184
 185 /*
 186  * bpf_prog should be a better name but it has been used in
 187  * linux/filter.h.
 188  */
 189 struct bpf_program {
 190         /* Index in elf obj file, for relocation use. */
 191         int idx;
 192         char *name;
 193         int prog_ifindex;
 194         char *section_name;
 195         /* section_name with / replaced by _; makes recursive pinning
 196          * in bpf_object__pin_programs easier
 197          */
 198         char *pin_name;
 199         struct bpf_insn *insns;
 200         size_t insns_cnt, main_prog_cnt;
 201         enum bpf_prog_type type;
 202
 203         struct reloc_desc *reloc_desc;
 204         int nr_reloc;
 205         int log_level;
 206
 207         struct {
 208                 int nr;
 209                 int *fds;
 210         } instances;
 211         bpf_program_prep_t preprocessor;
 212
 213         struct bpf_object *obj;
 214         void *priv;
 215         bpf_program_clear_priv_t clear_priv;
 216
 217         enum bpf_attach_type expected_attach_type;
 218         __u32 attach_btf_id;
 219         __u32 attach_prog_fd;
 220         void *func_info;
 221         __u32 func_info_rec_size;
 222         __u32 func_info_cnt;
 223
 224         struct bpf_capabilities *caps;
 225
 226         void *line_info;
 227         __u32 line_info_rec_size;
 228         __u32 line_info_cnt;
 229         __u32 prog_flags;
 230 };
 231
 232 #define DATA_SEC ".data"
 233 #define BSS_SEC ".bss"
 234 #define RODATA_SEC ".rodata"
 235 #define KCONFIG_SEC ".kconfig"
 236
 237 enum libbpf_map_type {
 238         LIBBPF_MAP_UNSPEC,
 239         LIBBPF_MAP_DATA,
 240         LIBBPF_MAP_BSS,
 241         LIBBPF_MAP_RODATA,
 242         LIBBPF_MAP_KCONFIG,
 243 };
 244
 245 static const char * const libbpf_type_to_btf_name[] = {
 246         [LIBBPF_MAP_DATA]       = DATA_SEC,
 247         [LIBBPF_MAP_BSS]        = BSS_SEC,
 248         [LIBBPF_MAP_RODATA]     = RODATA_SEC,
 249         [LIBBPF_MAP_KCONFIG]    = KCONFIG_SEC,
 250 };
 251
 252 struct bpf_map {
 253         char *name;
 254         int fd;
 255         int sec_idx;
 256         size_t sec_offset;
 257         int map_ifindex;
 258         int inner_map_fd;
 259         struct bpf_map_def def;
 260         __u32 btf_key_type_id;
 261         __u32 btf_value_type_id;
 262         void *priv;
 263         bpf_map_clear_priv_t clear_priv;
 264         enum libbpf_map_type libbpf_type;
 265         void *mmaped;
 266         char *pin_path;
 267         bool pinned;
 268         bool reused;
 269 };
 270
 271 enum extern_type {
 272         EXT_UNKNOWN,
 273         EXT_CHAR,
 274         EXT_BOOL,
 275         EXT_INT,
 276         EXT_TRISTATE,
 277         EXT_CHAR_ARR,
 278 };
 279
 280 struct extern_desc {
 281         const char *name;
 282         int sym_idx;
 283         int btf_id;
 284         enum extern_type type;
 285         int sz;
 286         int align;
 287         int data_off;
 288         bool is_signed;
 289         bool is_weak;
 290         bool is_set;
 291 };
 292
 293 static LIST_HEAD(bpf_objects_list);
 294
 295 struct bpf_object {
 296         char name[BPF_OBJ_NAME_LEN];
 297         char license[64];
 298         __u32 kern_version;
 299
 300         struct bpf_program *programs;
 301         size_t nr_programs;
 302         struct bpf_map *maps;
 303         size_t nr_maps;
 304         size_t maps_cap;
 305
 306         char *kconfig;
 307         struct extern_desc *externs;
 308         int nr_extern;
 309         int kconfig_map_idx;
 310
 311         bool loaded;
 312         bool has_pseudo_calls;
 313         bool relaxed_core_relocs;
 314
 315         /*
 316          * Information when doing elf related work. Only valid if fd
 317          * is valid.
 318          */
 319         struct {
 320                 int fd;
 321                 const void *obj_buf;
 322                 size_t obj_buf_sz;
 323                 Elf *elf;
 324                 GElf_Ehdr ehdr;
 325                 Elf_Data *symbols;
 326                 Elf_Data *data;
 327                 Elf_Data *rodata;
 328                 Elf_Data *bss;
 329                 size_t strtabidx;
 330                 struct {
 331                         GElf_Shdr shdr;
 332                         Elf_Data *data;
 333                 } *reloc_sects;
 334                 int nr_reloc_sects;
 335                 int maps_shndx;
 336                 int btf_maps_shndx;
 337                 int text_shndx;
 338                 int symbols_shndx;
 339                 int data_shndx;
 340                 int rodata_shndx;
 341                 int bss_shndx;
 342         } efile;
 343         /*
 344          * All loaded bpf_object is linked in a list, which is
 345          * hidden to caller. bpf_objects__<func> handlers deal with
 346          * all objects.
 347          */
 348         struct list_head list;
 349
 350         struct btf *btf;
 351         struct btf_ext *btf_ext;
 352
 353         void *priv;
 354         bpf_object_clear_priv_t clear_priv;
 355
 356         struct bpf_capabilities caps;
 357
 358         char path[];
 359 };
 360 #define obj_elf_valid(o)        ((o)->efile.elf)
 361
 362 void bpf_program__unload(struct bpf_program *prog)
 363 {
 364         int i;
 365
 366         if (!prog)
 367                 return;
 368
 369         /*
 370          * If the object is opened but the program was never loaded,
 371          * it is possible that prog->instances.nr == -1.
 372          */
 373         if (prog->instances.nr > 0) {
 374                 for (i = 0; i < prog->instances.nr; i++)
 375                         zclose(prog->instances.fds[i]);
 376         } else if (prog->instances.nr != -1) {
 377                 pr_warn("Internal error: instances.nr is %d\n",
 378                         prog->instances.nr);
 379         }
 380
 381         prog->instances.nr = -1;
 382         zfree(&prog->instances.fds);
 383
 384         zfree(&prog->func_info);
 385         zfree(&prog->line_info);
 386 }
 387
 388 static void bpf_program__exit(struct bpf_program *prog)
 389 {
 390         if (!prog)
 391                 return;
 392
 393         if (prog->clear_priv)
 394                 prog->clear_priv(prog, prog->priv);
 395
 396         prog->priv = NULL;
 397         prog->clear_priv = NULL;
 398
 399         bpf_program__unload(prog);
 400         zfree(&prog->name);
 401         zfree(&prog->section_name);
 402         zfree(&prog->pin_name);
 403         zfree(&prog->insns);
 404         zfree(&prog->reloc_desc);
 405
 406         prog->nr_reloc = 0;
 407         prog->insns_cnt = 0;
 408         prog->idx = -1;
 409 }
 410
 411 static char *__bpf_program__pin_name(struct bpf_program *prog)
 412 {
 413         char *name, *p;
 414
 415         name = p = strdup(prog->section_name);
 416         while ((p = strchr(p, '/')))
 417                 *p = '_';
 418
 419         return name;
 420 }
 421
 422 static int
 423 bpf_program__init(void *data, size_t size, char *section_name, int idx,
 424                   struct bpf_program *prog)
 425 {
 426         const size_t bpf_insn_sz = sizeof(struct bpf_insn);
 427
 428         if (size == 0 || size % bpf_insn_sz) {
 429                 pr_warn("corrupted section '%s', size: %zu\n",
 430                         section_name, size);
 431                 return -EINVAL;
 432         }
 433
 434         memset(prog, 0, sizeof(*prog));
 435
 436         prog->section_name = strdup(section_name);
 437         if (!prog->section_name) {
 438                 pr_warn("failed to alloc name for prog under section(%d) %s\n",
 439                         idx, section_name);
 440                 goto errout;
 441         }
 442
 443         prog->pin_name = __bpf_program__pin_name(prog);
 444         if (!prog->pin_name) {
 445                 pr_warn("failed to alloc pin name for prog under section(%d) %s\n",
 446                         idx, section_name);
 447                 goto errout;
 448         }
 449
 450         prog->insns = malloc(size);
 451         if (!prog->insns) {
 452                 pr_warn("failed to alloc insns for prog under section %s\n",
 453                         section_name);
 454                 goto errout;
 455         }
 456         prog->insns_cnt = size / bpf_insn_sz;
 457         memcpy(prog->insns, data, size);
 458         prog->idx = idx;
 459         prog->instances.fds = NULL;
 460         prog->instances.nr = -1;
 461         prog->type = BPF_PROG_TYPE_UNSPEC;
 462
 463         return 0;
 464 errout:
 465         bpf_program__exit(prog);
 466         return -ENOMEM;
 467 }
 468
 469 static int
 470 bpf_object__add_program(struct bpf_object *obj, void *data, size_t size,
 471                         char *section_name, int idx)
 472 {
 473         struct bpf_program prog, *progs;
 474         int nr_progs, err;
 475
 476         err = bpf_program__init(data, size, section_name, idx, &prog);
 477         if (err)
 478                 return err;
 479
 480         prog.caps = &obj->caps;
 481         progs = obj->programs;
 482         nr_progs = obj->nr_programs;
 483
 484         progs = reallocarray(progs, nr_progs + 1, sizeof(progs[0]));
 485         if (!progs) {
 486                 /*
 487                  * In this case the original obj->programs
 488                  * is still valid, so don't need special treat for
 489                  * bpf_close_object().
 490                  */
 491                 pr_warn("failed to alloc a new program under section '%s'\n",
 492                         section_name);
 493                 bpf_program__exit(&prog);
 494                 return -ENOMEM;
 495         }
 496
 497         pr_debug("found program %s\n", prog.section_name);
 498         obj->programs = progs;
 499         obj->nr_programs = nr_progs + 1;
 500         prog.obj = obj;
 501         progs[nr_progs] = prog;
 502         return 0;
 503 }
 504
 505 static int
 506 bpf_object__init_prog_names(struct bpf_object *obj)
 507 {
 508         Elf_Data *symbols = obj->efile.symbols;
 509         struct bpf_program *prog;
 510         size_t pi, si;
 511
 512         for (pi = 0; pi < obj->nr_programs; pi++) {
 513                 const char *name = NULL;
 514
 515                 prog = &obj->programs[pi];
 516
 517                 for (si = 0; si < symbols->d_size / sizeof(GElf_Sym) && !name;
 518                      si++) {
 519                         GElf_Sym sym;
 520
 521                         if (!gelf_getsym(symbols, si, &sym))
 522                                 continue;
 523                         if (sym.st_shndx != prog->idx)
 524                                 continue;
 525                         if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL)
 526                                 continue;
 527
 528                         name = elf_strptr(obj->efile.elf,
 529                                           obj->efile.strtabidx,
 530                                           sym.st_name);
 531                         if (!name) {
 532                                 pr_warn("failed to get sym name string for prog %s\n",
 533                                         prog->section_name);
 534                                 return -LIBBPF_ERRNO__LIBELF;
 535                         }
 536                 }
 537
 538                 if (!name && prog->idx == obj->efile.text_shndx)
 539                         name = ".text";
 540
 541                 if (!name) {
 542                         pr_warn("failed to find sym for prog %s\n",
 543                                 prog->section_name);
 544                         return -EINVAL;
 545                 }
 546
 547                 prog->name = strdup(name);
 548                 if (!prog->name) {
 549                         pr_warn("failed to allocate memory for prog sym %s\n",
 550                                 name);
 551                         return -ENOMEM;
 552                 }
 553         }
 554
 555         return 0;
 556 }
 557
 558 static __u32 get_kernel_version(void)
 559 {
 560         __u32 major, minor, patch;
 561         struct utsname info;
 562
 563         uname(&info);
 564         if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3)
 565                 return 0;
 566         return KERNEL_VERSION(major, minor, patch);
 567 }
 568
 569 static struct bpf_object *bpf_object__new(const char *path,
 570                                           const void *obj_buf,
 571                                           size_t obj_buf_sz,
 572                                           const char *obj_name)
 573 {
 574         struct bpf_object *obj;
 575         char *end;
 576
 577         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
 578         if (!obj) {
 579                 pr_warn("alloc memory failed for %s\n", path);
 580                 return ERR_PTR(-ENOMEM);
 581         }
 582
 583         strcpy(obj->path, path);
 584         if (obj_name) {
 585                 strncpy(obj->name, obj_name, sizeof(obj->name) - 1);
 586                 obj->name[sizeof(obj->name) - 1] = 0;
 587         } else {
 588                 /* Using basename() GNU version which doesn't modify arg. */
 589                 strncpy(obj->name, basename((void *)path),
 590                         sizeof(obj->name) - 1);
 591                 end = strchr(obj->name, '.');
 592                 if (end)
 593                         *end = 0;
 594         }
 595
 596         obj->efile.fd = -1;
 597         /*
 598          * Caller of this function should also call
 599          * bpf_object__elf_finish() after data collection to return
 600          * obj_buf to user. If not, we should duplicate the buffer to
 601          * avoid user freeing them before elf finish.
 602          */
 603         obj->efile.obj_buf = obj_buf;
 604         obj->efile.obj_buf_sz = obj_buf_sz;
 605         obj->efile.maps_shndx = -1;
 606         obj->efile.btf_maps_shndx = -1;
 607         obj->efile.data_shndx = -1;
 608         obj->efile.rodata_shndx = -1;
 609         obj->efile.bss_shndx = -1;
 610         obj->kconfig_map_idx = -1;
 611
 612         obj->kern_version = get_kernel_version();
 613         obj->loaded = false;
 614
 615         INIT_LIST_HEAD(&obj->list);
 616         list_add(&obj->list, &bpf_objects_list);
 617         return obj;
 618 }
 619
 620 static void bpf_object__elf_finish(struct bpf_object *obj)
 621 {
 622         if (!obj_elf_valid(obj))
 623                 return;
 624
 625         if (obj->efile.elf) {
 626                 elf_end(obj->efile.elf);
 627                 obj->efile.elf = NULL;
 628         }
 629         obj->efile.symbols = NULL;
 630         obj->efile.data = NULL;
 631         obj->efile.rodata = NULL;
 632         obj->efile.bss = NULL;
 633
 634         zfree(&obj->efile.reloc_sects);
 635         obj->efile.nr_reloc_sects = 0;
 636         zclose(obj->efile.fd);
 637         obj->efile.obj_buf = NULL;
 638         obj->efile.obj_buf_sz = 0;
 639 }
 640
 641 static int bpf_object__elf_init(struct bpf_object *obj)
 642 {
 643         int err = 0;
 644         GElf_Ehdr *ep;
 645
 646         if (obj_elf_valid(obj)) {
 647                 pr_warn("elf init: internal error\n");
 648                 return -LIBBPF_ERRNO__LIBELF;
 649         }
 650
 651         if (obj->efile.obj_buf_sz > 0) {
 652                 /*
 653                  * obj_buf should have been validated by
 654                  * bpf_object__open_buffer().
 655                  */
 656                 obj->efile.elf = elf_memory((char *)obj->efile.obj_buf,
 657                                             obj->efile.obj_buf_sz);
 658         } else {
 659                 obj->efile.fd = open(obj->path, O_RDONLY);
 660                 if (obj->efile.fd < 0) {
 661                         char errmsg[STRERR_BUFSIZE], *cp;
 662
 663                         err = -errno;
 664                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
 665                         pr_warn("failed to open %s: %s\n", obj->path, cp);
 666                         return err;
 667                 }
 668
 669                 obj->efile.elf = elf_begin(obj->efile.fd,
 670                                            LIBBPF_ELF_C_READ_MMAP, NULL);
 671         }
 672
 673         if (!obj->efile.elf) {
 674                 pr_warn("failed to open %s as ELF file\n", obj->path);
 675                 err = -LIBBPF_ERRNO__LIBELF;
 676                 goto errout;
 677         }
 678
 679         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
 680                 pr_warn("failed to get EHDR from %s\n", obj->path);
 681                 err = -LIBBPF_ERRNO__FORMAT;
 682                 goto errout;
 683         }
 684         ep = &obj->efile.ehdr;
 685
 686         /* Old LLVM set e_machine to EM_NONE */
 687         if (ep->e_type != ET_REL ||
 688             (ep->e_machine && ep->e_machine != EM_BPF)) {
 689                 pr_warn("%s is not an eBPF object file\n", obj->path);
 690                 err = -LIBBPF_ERRNO__FORMAT;
 691                 goto errout;
 692         }
 693
 694         return 0;
 695 errout:
 696         bpf_object__elf_finish(obj);
 697         return err;
 698 }
 699
 700 static int bpf_object__check_endianness(struct bpf_object *obj)
 701 {
 702 #if __BYTE_ORDER == __LITTLE_ENDIAN
 703         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
 704                 return 0;
 705 #elif __BYTE_ORDER == __BIG_ENDIAN
 706         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
 707                 return 0;
 708 #else
 709 # error "Unrecognized __BYTE_ORDER__"
 710 #endif
 711         pr_warn("endianness mismatch.\n");
 712         return -LIBBPF_ERRNO__ENDIAN;
 713 }
 714
 715 static int
 716 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
 717 {
 718         memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
 719         pr_debug("license of %s is %s\n", obj->path, obj->license);
 720         return 0;
 721 }
 722
 723 static int
 724 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
 725 {
 726         __u32 kver;
 727
 728         if (size != sizeof(kver)) {
 729                 pr_warn("invalid kver section in %s\n", obj->path);
 730                 return -LIBBPF_ERRNO__FORMAT;
 731         }
 732         memcpy(&kver, data, sizeof(kver));
 733         obj->kern_version = kver;
 734         pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
 735         return 0;
 736 }
 737
 738 static int compare_bpf_map(const void *_a, const void *_b)
 739 {
 740         const struct bpf_map *a = _a;
 741         const struct bpf_map *b = _b;
 742
 743         if (a->sec_idx != b->sec_idx)
 744                 return a->sec_idx - b->sec_idx;
 745         return a->sec_offset - b->sec_offset;
 746 }
 747
 748 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
 749 {
 750         if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
 751             type == BPF_MAP_TYPE_HASH_OF_MAPS)
 752                 return true;
 753         return false;
 754 }
 755
 756 static int bpf_object_search_section_size(const struct bpf_object *obj,
 757                                           const char *name, size_t *d_size)
 758 {
 759         const GElf_Ehdr *ep = &obj->efile.ehdr;
 760         Elf *elf = obj->efile.elf;
 761         Elf_Scn *scn = NULL;
 762         int idx = 0;
 763
 764         while ((scn = elf_nextscn(elf, scn)) != NULL) {
 765                 const char *sec_name;
 766                 Elf_Data *data;
 767                 GElf_Shdr sh;
 768
 769                 idx++;
 770                 if (gelf_getshdr(scn, &sh) != &sh) {
 771                         pr_warn("failed to get section(%d) header from %s\n",
 772                                 idx, obj->path);
 773                         return -EIO;
 774                 }
 775
 776                 sec_name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
 777                 if (!sec_name) {
 778                         pr_warn("failed to get section(%d) name from %s\n",
 779                                 idx, obj->path);
 780                         return -EIO;
 781                 }
 782
 783                 if (strcmp(name, sec_name))
 784                         continue;
 785
 786                 data = elf_getdata(scn, 0);
 787                 if (!data) {
 788                         pr_warn("failed to get section(%d) data from %s(%s)\n",
 789                                 idx, name, obj->path);
 790                         return -EIO;
 791                 }
 792
 793                 *d_size = data->d_size;
 794                 return 0;
 795         }
 796
 797         return -ENOENT;
 798 }
 799
 800 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
 801                              __u32 *size)
 802 {
 803         int ret = -ENOENT;
 804         size_t d_size;
 805
 806         *size = 0;
 807         if (!name) {
 808                 return -EINVAL;
 809         } else if (!strcmp(name, DATA_SEC)) {
 810                 if (obj->efile.data)
 811                         *size = obj->efile.data->d_size;
 812         } else if (!strcmp(name, BSS_SEC)) {
 813                 if (obj->efile.bss)
 814                         *size = obj->efile.bss->d_size;
 815         } else if (!strcmp(name, RODATA_SEC)) {
 816                 if (obj->efile.rodata)
 817                         *size = obj->efile.rodata->d_size;
 818         } else {
 819                 ret = bpf_object_search_section_size(obj, name, &d_size);
 820                 if (!ret)
 821                         *size = d_size;
 822         }
 823
 824         return *size ? 0 : ret;
 825 }
 826
 827 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
 828                                 __u32 *off)
 829 {
 830         Elf_Data *symbols = obj->efile.symbols;
 831         const char *sname;
 832         size_t si;
 833
 834         if (!name || !off)
 835                 return -EINVAL;
 836
 837         for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
 838                 GElf_Sym sym;
 839
 840                 if (!gelf_getsym(symbols, si, &sym))
 841                         continue;
 842                 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
 843                     GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
 844                         continue;
 845
 846                 sname = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
 847                                    sym.st_name);
 848                 if (!sname) {
 849                         pr_warn("failed to get sym name string for var %s\n",
 850                                 name);
 851                         return -EIO;
 852                 }
 853                 if (strcmp(name, sname) == 0) {
 854                         *off = sym.st_value;
 855                         return 0;
 856                 }
 857         }
 858
 859         return -ENOENT;
 860 }
 861
 862 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
 863 {
 864         struct bpf_map *new_maps;
 865         size_t new_cap;
 866         int i;
 867
 868         if (obj->nr_maps < obj->maps_cap)
 869                 return &obj->maps[obj->nr_maps++];
 870
 871         new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
 872         new_maps = realloc(obj->maps, new_cap * sizeof(*obj->maps));
 873         if (!new_maps) {
 874                 pr_warn("alloc maps for object failed\n");
 875                 return ERR_PTR(-ENOMEM);
 876         }
 877
 878         obj->maps_cap = new_cap;
 879         obj->maps = new_maps;
 880
 881         /* zero out new maps */
 882         memset(obj->maps + obj->nr_maps, 0,
 883                (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
 884         /*
 885          * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
 886          * when failure (zclose won't close negative fd)).
 887          */
 888         for (i = obj->nr_maps; i < obj->maps_cap; i++) {
 889                 obj->maps[i].fd = -1;
 890                 obj->maps[i].inner_map_fd = -1;
 891         }
 892
 893         return &obj->maps[obj->nr_maps++];
 894 }
 895
 896 static size_t bpf_map_mmap_sz(const struct bpf_map *map)
 897 {
 898         long page_sz = sysconf(_SC_PAGE_SIZE);
 899         size_t map_sz;
 900
 901         map_sz = roundup(map->def.value_size, 8) * map->def.max_entries;
 902         map_sz = roundup(map_sz, page_sz);
 903         return map_sz;
 904 }
 905
 906 static char *internal_map_name(struct bpf_object *obj,
 907                                enum libbpf_map_type type)
 908 {
 909         char map_name[BPF_OBJ_NAME_LEN];
 910         const char *sfx = libbpf_type_to_btf_name[type];
 911         int sfx_len = max((size_t)7, strlen(sfx));
 912         int pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1,
 913                           strlen(obj->name));
 914
 915         snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
 916                  sfx_len, libbpf_type_to_btf_name[type]);
 917
 918         return strdup(map_name);
 919 }
 920
 921 static int
 922 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
 923                               int sec_idx, void *data, size_t data_sz)
 924 {
 925         struct bpf_map_def *def;
 926         struct bpf_map *map;
 927         int err;
 928
 929         map = bpf_object__add_map(obj);
 930         if (IS_ERR(map))
 931                 return PTR_ERR(map);
 932
 933         map->libbpf_type = type;
 934         map->sec_idx = sec_idx;
 935         map->sec_offset = 0;
 936         map->name = internal_map_name(obj, type);
 937         if (!map->name) {
 938                 pr_warn("failed to alloc map name\n");
 939                 return -ENOMEM;
 940         }
 941
 942         def = &map->def;
 943         def->type = BPF_MAP_TYPE_ARRAY;
 944         def->key_size = sizeof(int);
 945         def->value_size = data_sz;
 946         def->max_entries = 1;
 947         def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
 948                          ? BPF_F_RDONLY_PROG : 0;
 949         def->map_flags |= BPF_F_MMAPABLE;
 950
 951         pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
 952                  map->name, map->sec_idx, map->sec_offset, def->map_flags);
 953
 954         map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE,
 955                            MAP_SHARED | MAP_ANONYMOUS, -1, 0);
 956         if (map->mmaped == MAP_FAILED) {
 957                 err = -errno;
 958                 map->mmaped = NULL;
 959                 pr_warn("failed to alloc map '%s' content buffer: %d\n",
 960                         map->name, err);
 961                 zfree(&map->name);
 962                 return err;
 963         }
 964
 965         if (data)
 966                 memcpy(map->mmaped, data, data_sz);
 967
 968         pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
 969         return 0;
 970 }
 971
 972 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
 973 {
 974         int err;
 975
 976         /*
 977          * Populate obj->maps with libbpf internal maps.
 978          */
 979         if (obj->efile.data_shndx >= 0) {
 980                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
 981                                                     obj->efile.data_shndx,
 982                                                     obj->efile.data->d_buf,
 983                                                     obj->efile.data->d_size);
 984                 if (err)
 985                         return err;
 986         }
 987         if (obj->efile.rodata_shndx >= 0) {
 988                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
 989                                                     obj->efile.rodata_shndx,
 990                                                     obj->efile.rodata->d_buf,
 991                                                     obj->efile.rodata->d_size);
 992                 if (err)
 993                         return err;
 994         }
 995         if (obj->efile.bss_shndx >= 0) {
 996                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
 997                                                     obj->efile.bss_shndx,
 998                                                     NULL,
 999                                                     obj->efile.bss->d_size);
1000                 if (err)
1001                         return err;
1002         }
1003         return 0;
1004 }
1005
1006
1007 static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
1008                                                const void *name)
1009 {
1010         int i;
1011
1012         for (i = 0; i < obj->nr_extern; i++) {
1013                 if (strcmp(obj->externs[i].name, name) == 0)
1014                         return &obj->externs[i];
1015         }
1016         return NULL;
1017 }
1018
1019 static int set_ext_value_tri(struct extern_desc *ext, void *ext_val,
1020                              char value)
1021 {
1022         switch (ext->type) {
1023         case EXT_BOOL:
1024                 if (value == 'm') {
1025                         pr_warn("extern %s=%c should be tristate or char\n",
1026                                 ext->name, value);
1027                         return -EINVAL;
1028                 }
1029                 *(bool *)ext_val = value == 'y' ? true : false;
1030                 break;
1031         case EXT_TRISTATE:
1032                 if (value == 'y')
1033                         *(enum libbpf_tristate *)ext_val = TRI_YES;
1034                 else if (value == 'm')
1035                         *(enum libbpf_tristate *)ext_val = TRI_MODULE;
1036                 else /* value == 'n' */
1037                         *(enum libbpf_tristate *)ext_val = TRI_NO;
1038                 break;
1039         case EXT_CHAR:
1040                 *(char *)ext_val = value;
1041                 break;
1042         case EXT_UNKNOWN:
1043         case EXT_INT:
1044         case EXT_CHAR_ARR:
1045         default:
1046                 pr_warn("extern %s=%c should be bool, tristate, or char\n",
1047                         ext->name, value);
1048                 return -EINVAL;
1049         }
1050         ext->is_set = true;
1051         return 0;
1052 }
1053
1054 static int set_ext_value_str(struct extern_desc *ext, char *ext_val,
1055                              const char *value)
1056 {
1057         size_t len;
1058
1059         if (ext->type != EXT_CHAR_ARR) {
1060                 pr_warn("extern %s=%s should char array\n", ext->name, value);
1061                 return -EINVAL;
1062         }
1063
1064         len = strlen(value);
1065         if (value[len - 1] != '"') {
1066                 pr_warn("extern '%s': invalid string config '%s'\n",
1067                         ext->name, value);
1068                 return -EINVAL;
1069         }
1070
1071         /* strip quotes */
1072         len -= 2;
1073         if (len >= ext->sz) {
1074                 pr_warn("extern '%s': long string config %s of (%zu bytes) truncated to %d bytes\n",
1075                         ext->name, value, len, ext->sz - 1);
1076                 len = ext->sz - 1;
1077         }
1078         memcpy(ext_val, value + 1, len);
1079         ext_val[len] = '\0';
1080         ext->is_set = true;
1081         return 0;
1082 }
1083
1084 static int parse_u64(const char *value, __u64 *res)
1085 {
1086         char *value_end;
1087         int err;
1088
1089         errno = 0;
1090         *res = strtoull(value, &value_end, 0);
1091         if (errno) {
1092                 err = -errno;
1093                 pr_warn("failed to parse '%s' as integer: %d\n", value, err);
1094                 return err;
1095         }
1096         if (*value_end) {
1097                 pr_warn("failed to parse '%s' as integer completely\n", value);
1098                 return -EINVAL;
1099         }
1100         return 0;
1101 }
1102
1103 static bool is_ext_value_in_range(const struct extern_desc *ext, __u64 v)
1104 {
1105         int bit_sz = ext->sz * 8;
1106
1107         if (ext->sz == 8)
1108                 return true;
1109
1110         /* Validate that value stored in u64 fits in integer of `ext->sz`
1111          * bytes size without any loss of information. If the target integer
1112          * is signed, we rely on the following limits of integer type of
1113          * Y bits and subsequent transformation:
1114          *
1115          *     -2^(Y-1) <= X           <= 2^(Y-1) - 1
1116          *            0 <= X + 2^(Y-1) <= 2^Y - 1
1117          *            0 <= X + 2^(Y-1) <  2^Y
1118          *
1119          *  For unsigned target integer, check that all the (64 - Y) bits are
1120          *  zero.
1121          */
1122         if (ext->is_signed)
1123                 return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
1124         else
1125                 return (v >> bit_sz) == 0;
1126 }
1127
1128 static int set_ext_value_num(struct extern_desc *ext, void *ext_val,
1129                              __u64 value)
1130 {
1131         if (ext->type != EXT_INT && ext->type != EXT_CHAR) {
1132                 pr_warn("extern %s=%llu should be integer\n",
1133                         ext->name, (unsigned long long)value);
1134                 return -EINVAL;
1135         }
1136         if (!is_ext_value_in_range(ext, value)) {
1137                 pr_warn("extern %s=%llu value doesn't fit in %d bytes\n",
1138                         ext->name, (unsigned long long)value, ext->sz);
1139                 return -ERANGE;
1140         }
1141         switch (ext->sz) {
1142                 case 1: *(__u8 *)ext_val = value; break;
1143                 case 2: *(__u16 *)ext_val = value; break;
1144                 case 4: *(__u32 *)ext_val = value; break;
1145                 case 8: *(__u64 *)ext_val = value; break;
1146                 default:
1147                         return -EINVAL;
1148         }
1149         ext->is_set = true;
1150         return 0;
1151 }
1152
1153 static int bpf_object__process_kconfig_line(struct bpf_object *obj,
1154                                             char *buf, void *data)
1155 {
1156         struct extern_desc *ext;
1157         char *sep, *value;
1158         int len, err = 0;
1159         void *ext_val;
1160         __u64 num;
1161
1162         if (strncmp(buf, "CONFIG_", 7))
1163                 return 0;
1164
1165         sep = strchr(buf, '=');
1166         if (!sep) {
1167                 pr_warn("failed to parse '%s': no separator\n", buf);
1168                 return -EINVAL;
1169         }
1170
1171         /* Trim ending '\n' */
1172         len = strlen(buf);
1173         if (buf[len - 1] == '\n')
1174                 buf[len - 1] = '\0';
1175         /* Split on '=' and ensure that a value is present. */
1176         *sep = '\0';
1177         if (!sep[1]) {
1178                 *sep = '=';
1179                 pr_warn("failed to parse '%s': no value\n", buf);
1180                 return -EINVAL;
1181         }
1182
1183         ext = find_extern_by_name(obj, buf);
1184         if (!ext || ext->is_set)
1185                 return 0;
1186
1187         ext_val = data + ext->data_off;
1188         value = sep + 1;
1189
1190         switch (*value) {
1191         case 'y': case 'n': case 'm':
1192                 err = set_ext_value_tri(ext, ext_val, *value);
1193                 break;
1194         case '"':
1195                 err = set_ext_value_str(ext, ext_val, value);
1196                 break;
1197         default:
1198                 /* assume integer */
1199                 err = parse_u64(value, &num);
1200                 if (err) {
1201                         pr_warn("extern %s=%s should be integer\n",
1202                                 ext->name, value);
1203                         return err;
1204                 }
1205                 err = set_ext_value_num(ext, ext_val, num);
1206                 break;
1207         }
1208         if (err)
1209                 return err;
1210         pr_debug("extern %s=%s\n", ext->name, value);
1211         return 0;
1212 }
1213
1214 static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
1215 {
1216         char buf[PATH_MAX];
1217         struct utsname uts;
1218         int len, err = 0;
1219         gzFile file;
1220
1221         uname(&uts);
1222         len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
1223         if (len < 0)
1224                 return -EINVAL;
1225         else if (len >= PATH_MAX)
1226                 return -ENAMETOOLONG;
1227
1228         /* gzopen also accepts uncompressed files. */
1229         file = gzopen(buf, "r");
1230         if (!file)
1231                 file = gzopen("/proc/config.gz", "r");
1232
1233         if (!file) {
1234                 pr_warn("failed to open system Kconfig\n");
1235                 return -ENOENT;
1236         }
1237
1238         while (gzgets(file, buf, sizeof(buf))) {
1239                 err = bpf_object__process_kconfig_line(obj, buf, data);
1240                 if (err) {
1241                         pr_warn("error parsing system Kconfig line '%s': %d\n",
1242                                 buf, err);
1243                         goto out;
1244                 }
1245         }
1246
1247 out:
1248         gzclose(file);
1249         return err;
1250 }
1251
1252 static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
1253                                         const char *config, void *data)
1254 {
1255         char buf[PATH_MAX];
1256         int err = 0;
1257         FILE *file;
1258
1259         file = fmemopen((void *)config, strlen(config), "r");
1260         if (!file) {
1261                 err = -errno;
1262                 pr_warn("failed to open in-memory Kconfig: %d\n", err);
1263                 return err;
1264         }
1265
1266         while (fgets(buf, sizeof(buf), file)) {
1267                 err = bpf_object__process_kconfig_line(obj, buf, data);
1268                 if (err) {
1269                         pr_warn("error parsing in-memory Kconfig line '%s': %d\n",
1270                                 buf, err);
1271                         break;
1272                 }
1273         }
1274
1275         fclose(file);
1276         return err;
1277 }
1278
1279 static int bpf_object__init_kconfig_map(struct bpf_object *obj)
1280 {
1281         struct extern_desc *last_ext;
1282         size_t map_sz;
1283         int err;
1284
1285         if (obj->nr_extern == 0)
1286                 return 0;
1287
1288         last_ext = &obj->externs[obj->nr_extern - 1];
1289         map_sz = last_ext->data_off + last_ext->sz;
1290
1291         err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
1292                                             obj->efile.symbols_shndx,
1293                                             NULL, map_sz);
1294         if (err)
1295                 return err;
1296
1297         obj->kconfig_map_idx = obj->nr_maps - 1;
1298
1299         return 0;
1300 }
1301
1302 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
1303 {
1304         Elf_Data *symbols = obj->efile.symbols;
1305         int i, map_def_sz = 0, nr_maps = 0, nr_syms;
1306         Elf_Data *data = NULL;
1307         Elf_Scn *scn;
1308
1309         if (obj->efile.maps_shndx < 0)
1310                 return 0;
1311
1312         if (!symbols)
1313                 return -EINVAL;
1314
1315         scn = elf_getscn(obj->efile.elf, obj->efile.maps_shndx);
1316         if (scn)
1317                 data = elf_getdata(scn, NULL);
1318         if (!scn || !data) {
1319                 pr_warn("failed to get Elf_Data from map section %d\n",
1320                         obj->efile.maps_shndx);
1321                 return -EINVAL;
1322         }
1323
1324         /*
1325          * Count number of maps. Each map has a name.
1326          * Array of maps is not supported: only the first element is
1327          * considered.
1328          *
1329          * TODO: Detect array of map and report error.
1330          */
1331         nr_syms = symbols->d_size / sizeof(GElf_Sym);
1332         for (i = 0; i < nr_syms; i++) {
1333                 GElf_Sym sym;
1334
1335                 if (!gelf_getsym(symbols, i, &sym))
1336                         continue;
1337                 if (sym.st_shndx != obj->efile.maps_shndx)
1338                         continue;
1339                 nr_maps++;
1340         }
1341         /* Assume equally sized map definitions */
1342         pr_debug("maps in %s: %d maps in %zd bytes\n",
1343                  obj->path, nr_maps, data->d_size);
1344
1345         if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) {
1346                 pr_warn("unable to determine map definition size section %s, %d maps in %zd bytes\n",
1347                         obj->path, nr_maps, data->d_size);
1348                 return -EINVAL;
1349         }
1350         map_def_sz = data->d_size / nr_maps;
1351
1352         /* Fill obj->maps using data in "maps" section.  */
1353         for (i = 0; i < nr_syms; i++) {
1354                 GElf_Sym sym;
1355                 const char *map_name;
1356                 struct bpf_map_def *def;
1357                 struct bpf_map *map;
1358
1359                 if (!gelf_getsym(symbols, i, &sym))
1360                         continue;
1361                 if (sym.st_shndx != obj->efile.maps_shndx)
1362                         continue;
1363
1364                 map = bpf_object__add_map(obj);
1365                 if (IS_ERR(map))
1366                         return PTR_ERR(map);
1367
1368                 map_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
1369                                       sym.st_name);
1370                 if (!map_name) {
1371                         pr_warn("failed to get map #%d name sym string for obj %s\n",
1372                                 i, obj->path);
1373                         return -LIBBPF_ERRNO__FORMAT;
1374                 }
1375
1376                 map->libbpf_type = LIBBPF_MAP_UNSPEC;
1377                 map->sec_idx = sym.st_shndx;
1378                 map->sec_offset = sym.st_value;
1379                 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
1380                          map_name, map->sec_idx, map->sec_offset);
1381                 if (sym.st_value + map_def_sz > data->d_size) {
1382                         pr_warn("corrupted maps section in %s: last map \"%s\" too small\n",
1383                                 obj->path, map_name);
1384                         return -EINVAL;
1385                 }
1386
1387                 map->name = strdup(map_name);
1388                 if (!map->name) {
1389                         pr_warn("failed to alloc map name\n");
1390                         return -ENOMEM;
1391                 }
1392                 pr_debug("map %d is \"%s\"\n", i, map->name);
1393                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
1394                 /*
1395                  * If the definition of the map in the object file fits in
1396                  * bpf_map_def, copy it.  Any extra fields in our version
1397                  * of bpf_map_def will default to zero as a result of the
1398                  * calloc above.
1399                  */
1400                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
1401                         memcpy(&map->def, def, map_def_sz);
1402                 } else {
1403                         /*
1404                          * Here the map structure being read is bigger than what
1405                          * we expect, truncate if the excess bits are all zero.
1406                          * If they are not zero, reject this map as
1407                          * incompatible.
1408                          */
1409                         char *b;
1410
1411                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
1412                              b < ((char *)def) + map_def_sz; b++) {
1413                                 if (*b != 0) {
1414                                         pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n",
1415                                                 obj->path, map_name);
1416                                         if (strict)
1417                                                 return -EINVAL;
1418                                 }
1419                         }
1420                         memcpy(&map->def, def, sizeof(struct bpf_map_def));
1421                 }
1422         }
1423         return 0;
1424 }
1425
1426 static const struct btf_type *
1427 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
1428 {
1429         const struct btf_type *t = btf__type_by_id(btf, id);
1430
1431         if (res_id)
1432                 *res_id = id;
1433
1434         while (btf_is_mod(t) || btf_is_typedef(t)) {
1435                 if (res_id)
1436                         *res_id = t->type;
1437                 t = btf__type_by_id(btf, t->type);
1438         }
1439
1440         return t;
1441 }
1442
1443 /*
1444  * Fetch integer attribute of BTF map definition. Such attributes are
1445  * represented using a pointer to an array, in which dimensionality of array
1446  * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
1447  * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
1448  * type definition, while using only sizeof(void *) space in ELF data section.
1449  */
1450 static bool get_map_field_int(const char *map_name, const struct btf *btf,
1451                               const struct btf_type *def,
1452                               const struct btf_member *m, __u32 *res)
1453 {
1454         const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
1455         const char *name = btf__name_by_offset(btf, m->name_off);
1456         const struct btf_array *arr_info;
1457         const struct btf_type *arr_t;
1458
1459         if (!btf_is_ptr(t)) {
1460                 pr_warn("map '%s': attr '%s': expected PTR, got %u.\n",
1461                         map_name, name, btf_kind(t));
1462                 return false;
1463         }
1464
1465         arr_t = btf__type_by_id(btf, t->type);
1466         if (!arr_t) {
1467                 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
1468                         map_name, name, t->type);
1469                 return false;
1470         }
1471         if (!btf_is_array(arr_t)) {
1472                 pr_warn("map '%s': attr '%s': expected ARRAY, got %u.\n",
1473                         map_name, name, btf_kind(arr_t));
1474                 return false;
1475         }
1476         arr_info = btf_array(arr_t);
1477         *res = arr_info->nelems;
1478         return true;
1479 }
1480
1481 static int build_map_pin_path(struct bpf_map *map, const char *path)
1482 {
1483         char buf[PATH_MAX];
1484         int err, len;
1485
1486         if (!path)
1487                 path = "/sys/fs/bpf";
1488
1489         len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map));
1490         if (len < 0)
1491                 return -EINVAL;
1492         else if (len >= PATH_MAX)
1493                 return -ENAMETOOLONG;
1494
1495         err = bpf_map__set_pin_path(map, buf);
1496         if (err)
1497                 return err;
1498
1499         return 0;
1500 }
1501
1502 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
1503                                          const struct btf_type *sec,
1504                                          int var_idx, int sec_idx,
1505                                          const Elf_Data *data, bool strict,
1506                                          const char *pin_root_path)
1507 {
1508         const struct btf_type *var, *def, *t;
1509         const struct btf_var_secinfo *vi;
1510         const struct btf_var *var_extra;
1511         const struct btf_member *m;
1512         const char *map_name;
1513         struct bpf_map *map;
1514         int vlen, i;
1515
1516         vi = btf_var_secinfos(sec) + var_idx;
1517         var = btf__type_by_id(obj->btf, vi->type);
1518         var_extra = btf_var(var);
1519         map_name = btf__name_by_offset(obj->btf, var->name_off);
1520         vlen = btf_vlen(var);
1521
1522         if (map_name == NULL || map_name[0] == '\0') {
1523                 pr_warn("map #%d: empty name.\n", var_idx);
1524                 return -EINVAL;
1525         }
1526         if ((__u64)vi->offset + vi->size > data->d_size) {
1527                 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
1528                 return -EINVAL;
1529         }
1530         if (!btf_is_var(var)) {
1531                 pr_warn("map '%s': unexpected var kind %u.\n",
1532                         map_name, btf_kind(var));
1533                 return -EINVAL;
1534         }
1535         if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
1536             var_extra->linkage != BTF_VAR_STATIC) {
1537                 pr_warn("map '%s': unsupported var linkage %u.\n",
1538                         map_name, var_extra->linkage);
1539                 return -EOPNOTSUPP;
1540         }
1541
1542         def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
1543         if (!btf_is_struct(def)) {
1544                 pr_warn("map '%s': unexpected def kind %u.\n",
1545                         map_name, btf_kind(var));
1546                 return -EINVAL;
1547         }
1548         if (def->size > vi->size) {
1549                 pr_warn("map '%s': invalid def size.\n", map_name);
1550                 return -EINVAL;
1551         }
1552
1553         map = bpf_object__add_map(obj);
1554         if (IS_ERR(map))
1555                 return PTR_ERR(map);
1556         map->name = strdup(map_name);
1557         if (!map->name) {
1558                 pr_warn("map '%s': failed to alloc map name.\n", map_name);
1559                 return -ENOMEM;
1560         }
1561         map->libbpf_type = LIBBPF_MAP_UNSPEC;
1562         map->def.type = BPF_MAP_TYPE_UNSPEC;
1563         map->sec_idx = sec_idx;
1564         map->sec_offset = vi->offset;
1565         pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
1566                  map_name, map->sec_idx, map->sec_offset);
1567
1568         vlen = btf_vlen(def);
1569         m = btf_members(def);
1570         for (i = 0; i < vlen; i++, m++) {
1571                 const char *name = btf__name_by_offset(obj->btf, m->name_off);
1572
1573                 if (!name) {
1574                         pr_warn("map '%s': invalid field #%d.\n", map_name, i);
1575                         return -EINVAL;
1576                 }
1577                 if (strcmp(name, "type") == 0) {
1578                         if (!get_map_field_int(map_name, obj->btf, def, m,
1579                                                &map->def.type))
1580                                 return -EINVAL;
1581                         pr_debug("map '%s': found type = %u.\n",
1582                                  map_name, map->def.type);
1583                 } else if (strcmp(name, "max_entries") == 0) {
1584                         if (!get_map_field_int(map_name, obj->btf, def, m,
1585                                                &map->def.max_entries))
1586                                 return -EINVAL;
1587                         pr_debug("map '%s': found max_entries = %u.\n",
1588                                  map_name, map->def.max_entries);
1589                 } else if (strcmp(name, "map_flags") == 0) {
1590                         if (!get_map_field_int(map_name, obj->btf, def, m,
1591                                                &map->def.map_flags))
1592                                 return -EINVAL;
1593                         pr_debug("map '%s': found map_flags = %u.\n",
1594                                  map_name, map->def.map_flags);
1595                 } else if (strcmp(name, "key_size") == 0) {
1596                         __u32 sz;
1597
1598                         if (!get_map_field_int(map_name, obj->btf, def, m,
1599                                                &sz))
1600                                 return -EINVAL;
1601                         pr_debug("map '%s': found key_size = %u.\n",
1602                                  map_name, sz);
1603                         if (map->def.key_size && map->def.key_size != sz) {
1604                                 pr_warn("map '%s': conflicting key size %u != %u.\n",
1605                                         map_name, map->def.key_size, sz);
1606                                 return -EINVAL;
1607                         }
1608                         map->def.key_size = sz;
1609                 } else if (strcmp(name, "key") == 0) {
1610                         __s64 sz;
1611
1612                         t = btf__type_by_id(obj->btf, m->type);
1613                         if (!t) {
1614                                 pr_warn("map '%s': key type [%d] not found.\n",
1615                                         map_name, m->type);
1616                                 return -EINVAL;
1617                         }
1618                         if (!btf_is_ptr(t)) {
1619                                 pr_warn("map '%s': key spec is not PTR: %u.\n",
1620                                         map_name, btf_kind(t));
1621                                 return -EINVAL;
1622                         }
1623                         sz = btf__resolve_size(obj->btf, t->type);
1624                         if (sz < 0) {
1625                                 pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
1626                                         map_name, t->type, (ssize_t)sz);
1627                                 return sz;
1628                         }
1629                         pr_debug("map '%s': found key [%u], sz = %zd.\n",
1630                                  map_name, t->type, (ssize_t)sz);
1631                         if (map->def.key_size && map->def.key_size != sz) {
1632                                 pr_warn("map '%s': conflicting key size %u != %zd.\n",
1633                                         map_name, map->def.key_size, (ssize_t)sz);
1634                                 return -EINVAL;
1635                         }
1636                         map->def.key_size = sz;
1637                         map->btf_key_type_id = t->type;
1638                 } else if (strcmp(name, "value_size") == 0) {
1639                         __u32 sz;
1640
1641                         if (!get_map_field_int(map_name, obj->btf, def, m,
1642                                                &sz))
1643                                 return -EINVAL;
1644                         pr_debug("map '%s': found value_size = %u.\n",
1645                                  map_name, sz);
1646                         if (map->def.value_size && map->def.value_size != sz) {
1647                                 pr_warn("map '%s': conflicting value size %u != %u.\n",
1648                                         map_name, map->def.value_size, sz);
1649                                 return -EINVAL;
1650                         }
1651                         map->def.value_size = sz;
1652                 } else if (strcmp(name, "value") == 0) {
1653                         __s64 sz;
1654
1655                         t = btf__type_by_id(obj->btf, m->type);
1656                         if (!t) {
1657                                 pr_warn("map '%s': value type [%d] not found.\n",
1658                                         map_name, m->type);
1659                                 return -EINVAL;
1660                         }
1661                         if (!btf_is_ptr(t)) {
1662                                 pr_warn("map '%s': value spec is not PTR: %u.\n",
1663                                         map_name, btf_kind(t));
1664                                 return -EINVAL;
1665                         }
1666                         sz = btf__resolve_size(obj->btf, t->type);
1667                         if (sz < 0) {
1668                                 pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
1669                                         map_name, t->type, (ssize_t)sz);
1670                                 return sz;
1671                         }
1672                         pr_debug("map '%s': found value [%u], sz = %zd.\n",
1673                                  map_name, t->type, (ssize_t)sz);
1674                         if (map->def.value_size && map->def.value_size != sz) {
1675                                 pr_warn("map '%s': conflicting value size %u != %zd.\n",
1676                                         map_name, map->def.value_size, (ssize_t)sz);
1677                                 return -EINVAL;
1678                         }
1679                         map->def.value_size = sz;
1680                         map->btf_value_type_id = t->type;
1681                 } else if (strcmp(name, "pinning") == 0) {
1682                         __u32 val;
1683                         int err;
1684
1685                         if (!get_map_field_int(map_name, obj->btf, def, m,
1686                                                &val))
1687                                 return -EINVAL;
1688                         pr_debug("map '%s': found pinning = %u.\n",
1689                                  map_name, val);
1690
1691                         if (val != LIBBPF_PIN_NONE &&
1692                             val != LIBBPF_PIN_BY_NAME) {
1693                                 pr_warn("map '%s': invalid pinning value %u.\n",
1694                                         map_name, val);
1695                                 return -EINVAL;
1696                         }
1697                         if (val == LIBBPF_PIN_BY_NAME) {
1698                                 err = build_map_pin_path(map, pin_root_path);
1699                                 if (err) {
1700                                         pr_warn("map '%s': couldn't build pin path.\n",
1701                                                 map_name);
1702                                         return err;
1703                                 }
1704                         }
1705                 } else {
1706                         if (strict) {
1707                                 pr_warn("map '%s': unknown field '%s'.\n",
1708                                         map_name, name);
1709                                 return -ENOTSUP;
1710                         }
1711                         pr_debug("map '%s': ignoring unknown field '%s'.\n",
1712                                  map_name, name);
1713                 }
1714         }
1715
1716         if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
1717                 pr_warn("map '%s': map type isn't specified.\n", map_name);
1718                 return -EINVAL;
1719         }
1720
1721         return 0;
1722 }
1723
1724 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
1725                                           const char *pin_root_path)
1726 {
1727         const struct btf_type *sec = NULL;
1728         int nr_types, i, vlen, err;
1729         const struct btf_type *t;
1730         const char *name;
1731         Elf_Data *data;
1732         Elf_Scn *scn;
1733
1734         if (obj->efile.btf_maps_shndx < 0)
1735                 return 0;
1736
1737         scn = elf_getscn(obj->efile.elf, obj->efile.btf_maps_shndx);
1738         if (scn)
1739                 data = elf_getdata(scn, NULL);
1740         if (!scn || !data) {
1741                 pr_warn("failed to get Elf_Data from map section %d (%s)\n",
1742                         obj->efile.maps_shndx, MAPS_ELF_SEC);
1743                 return -EINVAL;
1744         }
1745
1746         nr_types = btf__get_nr_types(obj->btf);
1747         for (i = 1; i <= nr_types; i++) {
1748                 t = btf__type_by_id(obj->btf, i);
1749                 if (!btf_is_datasec(t))
1750                         continue;
1751                 name = btf__name_by_offset(obj->btf, t->name_off);
1752                 if (strcmp(name, MAPS_ELF_SEC) == 0) {
1753                         sec = t;
1754                         break;
1755                 }
1756         }
1757
1758         if (!sec) {
1759                 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
1760                 return -ENOENT;
1761         }
1762
1763         vlen = btf_vlen(sec);
1764         for (i = 0; i < vlen; i++) {
1765                 err = bpf_object__init_user_btf_map(obj, sec, i,
1766                                                     obj->efile.btf_maps_shndx,
1767                                                     data, strict,
1768                                                     pin_root_path);
1769                 if (err)
1770                         return err;
1771         }
1772
1773         return 0;
1774 }
1775
1776 static int bpf_object__init_maps(struct bpf_object *obj,
1777                                  const struct bpf_object_open_opts *opts)
1778 {
1779         const char *pin_root_path;
1780         bool strict;
1781         int err;
1782
1783         strict = !OPTS_GET(opts, relaxed_maps, false);
1784         pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
1785
1786         err = bpf_object__init_user_maps(obj, strict);
1787         err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
1788         err = err ?: bpf_object__init_global_data_maps(obj);
1789         err = err ?: bpf_object__init_kconfig_map(obj);
1790         if (err)
1791                 return err;
1792
1793         if (obj->nr_maps) {
1794                 qsort(obj->maps, obj->nr_maps, sizeof(obj->maps[0]),
1795                       compare_bpf_map);
1796         }
1797         return 0;
1798 }
1799
1800 static bool section_have_execinstr(struct bpf_object *obj, int idx)
1801 {
1802         Elf_Scn *scn;
1803         GElf_Shdr sh;
1804
1805         scn = elf_getscn(obj->efile.elf, idx);
1806         if (!scn)
1807                 return false;
1808
1809         if (gelf_getshdr(scn, &sh) != &sh)
1810                 return false;
1811
1812         if (sh.sh_flags & SHF_EXECINSTR)
1813                 return true;
1814
1815         return false;
1816 }
1817
1818 static void bpf_object__sanitize_btf(struct bpf_object *obj)
1819 {
1820         bool has_datasec = obj->caps.btf_datasec;
1821         bool has_func = obj->caps.btf_func;
1822         struct btf *btf = obj->btf;
1823         struct btf_type *t;
1824         int i, j, vlen;
1825
1826         if (!obj->btf || (has_func && has_datasec))
1827                 return;
1828
1829         for (i = 1; i <= btf__get_nr_types(btf); i++) {
1830                 t = (struct btf_type *)btf__type_by_id(btf, i);
1831
1832                 if (!has_datasec && btf_is_var(t)) {
1833                         /* replace VAR with INT */
1834                         t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
1835                         /*
1836                          * using size = 1 is the safest choice, 4 will be too
1837                          * big and cause kernel BTF validation failure if
1838                          * original variable took less than 4 bytes
1839                          */
1840                         t->size = 1;
1841                         *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
1842                 } else if (!has_datasec && btf_is_datasec(t)) {
1843                         /* replace DATASEC with STRUCT */
1844                         const struct btf_var_secinfo *v = btf_var_secinfos(t);
1845                         struct btf_member *m = btf_members(t);
1846                         struct btf_type *vt;
1847                         char *name;
1848
1849                         name = (char *)btf__name_by_offset(btf, t->name_off);
1850                         while (*name) {
1851                                 if (*name == '.')
1852                                         *name = '_';
1853                                 name++;
1854                         }
1855
1856                         vlen = btf_vlen(t);
1857                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
1858                         for (j = 0; j < vlen; j++, v++, m++) {
1859                                 /* order of field assignments is important */
1860                                 m->offset = v->offset * 8;
1861                                 m->type = v->type;
1862                                 /* preserve variable name as member name */
1863                                 vt = (void *)btf__type_by_id(btf, v->type);
1864                                 m->name_off = vt->name_off;
1865                         }
1866                 } else if (!has_func && btf_is_func_proto(t)) {
1867                         /* replace FUNC_PROTO with ENUM */
1868                         vlen = btf_vlen(t);
1869                         t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
1870                         t->size = sizeof(__u32); /* kernel enforced */
1871                 } else if (!has_func && btf_is_func(t)) {
1872                         /* replace FUNC with TYPEDEF */
1873                         t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
1874                 }
1875         }
1876 }
1877
1878 static void bpf_object__sanitize_btf_ext(struct bpf_object *obj)
1879 {
1880         if (!obj->btf_ext)
1881                 return;
1882
1883         if (!obj->caps.btf_func) {
1884                 btf_ext__free(obj->btf_ext);
1885                 obj->btf_ext = NULL;
1886         }
1887 }
1888
1889 static bool bpf_object__is_btf_mandatory(const struct bpf_object *obj)
1890 {
1891         return obj->efile.btf_maps_shndx >= 0 ||
1892                obj->nr_extern > 0;
1893 }
1894
1895 static int bpf_object__init_btf(struct bpf_object *obj,
1896                                 Elf_Data *btf_data,
1897                                 Elf_Data *btf_ext_data)
1898 {
1899         bool btf_required = bpf_object__is_btf_mandatory(obj);
1900         int err = 0;
1901
1902         if (btf_data) {
1903                 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
1904                 if (IS_ERR(obj->btf)) {
1905                         pr_warn("Error loading ELF section %s: %d.\n",
1906                                 BTF_ELF_SEC, err);
1907                         goto out;
1908                 }
1909         }
1910         if (btf_ext_data) {
1911                 if (!obj->btf) {
1912                         pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
1913                                  BTF_EXT_ELF_SEC, BTF_ELF_SEC);
1914                         goto out;
1915                 }
1916                 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
1917                                             btf_ext_data->d_size);
1918                 if (IS_ERR(obj->btf_ext)) {
1919                         pr_warn("Error loading ELF section %s: %ld. Ignored and continue.\n",
1920                                 BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
1921                         obj->btf_ext = NULL;
1922                         goto out;
1923                 }
1924         }
1925 out:
1926         if (err || IS_ERR(obj->btf)) {
1927                 if (btf_required)
1928                         err = err ? : PTR_ERR(obj->btf);
1929                 else
1930                         err = 0;
1931                 if (!IS_ERR_OR_NULL(obj->btf))
1932                         btf__free(obj->btf);
1933                 obj->btf = NULL;
1934         }
1935         if (btf_required && !obj->btf) {
1936                 pr_warn("BTF is required, but is missing or corrupted.\n");
1937                 return err == 0 ? -ENOENT : err;
1938         }
1939         return 0;
1940 }
1941
1942 static int bpf_object__finalize_btf(struct bpf_object *obj)
1943 {
1944         int err;
1945
1946         if (!obj->btf)
1947                 return 0;
1948
1949         err = btf__finalize_data(obj, obj->btf);
1950         if (!err)
1951                 return 0;
1952
1953         pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err);
1954         btf__free(obj->btf);
1955         obj->btf = NULL;
1956         btf_ext__free(obj->btf_ext);
1957         obj->btf_ext = NULL;
1958
1959         if (bpf_object__is_btf_mandatory(obj)) {
1960                 pr_warn("BTF is required, but is missing or corrupted.\n");
1961                 return -ENOENT;
1962         }
1963         return 0;
1964 }
1965
1966 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
1967 {
1968         int err = 0;
1969
1970         if (!obj->btf)
1971                 return 0;
1972
1973         bpf_object__sanitize_btf(obj);
1974         bpf_object__sanitize_btf_ext(obj);
1975
1976         err = btf__load(obj->btf);
1977         if (err) {
1978                 pr_warn("Error loading %s into kernel: %d.\n",
1979                         BTF_ELF_SEC, err);
1980                 btf__free(obj->btf);
1981                 obj->btf = NULL;
1982                 /* btf_ext can't exist without btf, so free it as well */
1983                 if (obj->btf_ext) {
1984                         btf_ext__free(obj->btf_ext);
1985                         obj->btf_ext = NULL;
1986                 }
1987
1988                 if (bpf_object__is_btf_mandatory(obj))
1989                         return err;
1990         }
1991         return 0;
1992 }
1993
1994 static int bpf_object__elf_collect(struct bpf_object *obj)
1995 {
1996         Elf *elf = obj->efile.elf;
1997         GElf_Ehdr *ep = &obj->efile.ehdr;
1998         Elf_Data *btf_ext_data = NULL;
1999         Elf_Data *btf_data = NULL;
2000         Elf_Scn *scn = NULL;
2001         int idx = 0, err = 0;
2002
2003         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
2004         if (!elf_rawdata(elf_getscn(elf, ep->e_shstrndx), NULL)) {
2005                 pr_warn("failed to get e_shstrndx from %s\n", obj->path);
2006                 return -LIBBPF_ERRNO__FORMAT;
2007         }
2008
2009         while ((scn = elf_nextscn(elf, scn)) != NULL) {
2010                 char *name;
2011                 GElf_Shdr sh;
2012                 Elf_Data *data;
2013
2014                 idx++;
2015                 if (gelf_getshdr(scn, &sh) != &sh) {
2016                         pr_warn("failed to get section(%d) header from %s\n",
2017                                 idx, obj->path);
2018                         return -LIBBPF_ERRNO__FORMAT;
2019                 }
2020
2021                 name = elf_strptr(elf, ep->e_shstrndx, sh.sh_name);
2022                 if (!name) {
2023                         pr_warn("failed to get section(%d) name from %s\n",
2024                                 idx, obj->path);
2025                         return -LIBBPF_ERRNO__FORMAT;
2026                 }
2027
2028                 data = elf_getdata(scn, 0);
2029                 if (!data) {
2030                         pr_warn("failed to get section(%d) data from %s(%s)\n",
2031                                 idx, name, obj->path);
2032                         return -LIBBPF_ERRNO__FORMAT;
2033                 }
2034                 pr_debug("section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
2035                          idx, name, (unsigned long)data->d_size,
2036                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
2037                          (int)sh.sh_type);
2038
2039                 if (strcmp(name, "license") == 0) {
2040                         err = bpf_object__init_license(obj,
2041                                                        data->d_buf,
2042                                                        data->d_size);
2043                         if (err)
2044                                 return err;
2045                 } else if (strcmp(name, "version") == 0) {
2046                         err = bpf_object__init_kversion(obj,
2047                                                         data->d_buf,
2048                                                         data->d_size);
2049                         if (err)
2050                                 return err;
2051                 } else if (strcmp(name, "maps") == 0) {
2052                         obj->efile.maps_shndx = idx;
2053                 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
2054                         obj->efile.btf_maps_shndx = idx;
2055                 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
2056                         btf_data = data;
2057                 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
2058                         btf_ext_data = data;
2059                 } else if (sh.sh_type == SHT_SYMTAB) {
2060                         if (obj->efile.symbols) {
2061                                 pr_warn("bpf: multiple SYMTAB in %s\n",
2062                                         obj->path);
2063                                 return -LIBBPF_ERRNO__FORMAT;
2064                         }
2065                         obj->efile.symbols = data;
2066                         obj->efile.symbols_shndx = idx;
2067                         obj->efile.strtabidx = sh.sh_link;
2068                 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
2069                         if (sh.sh_flags & SHF_EXECINSTR) {
2070                                 if (strcmp(name, ".text") == 0)
2071                                         obj->efile.text_shndx = idx;
2072                                 err = bpf_object__add_program(obj, data->d_buf,
2073                                                               data->d_size,
2074                                                               name, idx);
2075                                 if (err) {
2076                                         char errmsg[STRERR_BUFSIZE];
2077                                         char *cp;
2078
2079                                         cp = libbpf_strerror_r(-err, errmsg,
2080                                                                sizeof(errmsg));
2081                                         pr_warn("failed to alloc program %s (%s): %s",
2082                                                 name, obj->path, cp);
2083                                         return err;
2084                                 }
2085                         } else if (strcmp(name, DATA_SEC) == 0) {
2086                                 obj->efile.data = data;
2087                                 obj->efile.data_shndx = idx;
2088                         } else if (strcmp(name, RODATA_SEC) == 0) {
2089                                 obj->efile.rodata = data;
2090                                 obj->efile.rodata_shndx = idx;
2091                         } else {
2092                                 pr_debug("skip section(%d) %s\n", idx, name);
2093                         }
2094                 } else if (sh.sh_type == SHT_REL) {
2095                         int nr_sects = obj->efile.nr_reloc_sects;
2096                         void *sects = obj->efile.reloc_sects;
2097                         int sec = sh.sh_info; /* points to other section */
2098
2099                         /* Only do relo for section with exec instructions */
2100                         if (!section_have_execinstr(obj, sec)) {
2101                                 pr_debug("skip relo %s(%d) for section(%d)\n",
2102                                          name, idx, sec);
2103                                 continue;
2104                         }
2105
2106                         sects = reallocarray(sects, nr_sects + 1,
2107                                              sizeof(*obj->efile.reloc_sects));
2108                         if (!sects) {
2109                                 pr_warn("reloc_sects realloc failed\n");
2110                                 return -ENOMEM;
2111                         }
2112
2113                         obj->efile.reloc_sects = sects;
2114                         obj->efile.nr_reloc_sects++;
2115
2116                         obj->efile.reloc_sects[nr_sects].shdr = sh;
2117                         obj->efile.reloc_sects[nr_sects].data = data;
2118                 } else if (sh.sh_type == SHT_NOBITS &&
2119                            strcmp(name, BSS_SEC) == 0) {
2120                         obj->efile.bss = data;
2121                         obj->efile.bss_shndx = idx;
2122                 } else {
2123                         pr_debug("skip section(%d) %s\n", idx, name);
2124                 }
2125         }
2126
2127         if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
2128                 pr_warn("Corrupted ELF file: index of strtab invalid\n");
2129                 return -LIBBPF_ERRNO__FORMAT;
2130         }
2131         return bpf_object__init_btf(obj, btf_data, btf_ext_data);
2132 }
2133
2134 static bool sym_is_extern(const GElf_Sym *sym)
2135 {
2136         int bind = GELF_ST_BIND(sym->st_info);
2137         /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
2138         return sym->st_shndx == SHN_UNDEF &&
2139                (bind == STB_GLOBAL || bind == STB_WEAK) &&
2140                GELF_ST_TYPE(sym->st_info) == STT_NOTYPE;
2141 }
2142
2143 static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
2144 {
2145         const struct btf_type *t;
2146         const char *var_name;
2147         int i, n;
2148
2149         if (!btf)
2150                 return -ESRCH;
2151
2152         n = btf__get_nr_types(btf);
2153         for (i = 1; i <= n; i++) {
2154                 t = btf__type_by_id(btf, i);
2155
2156                 if (!btf_is_var(t))
2157                         continue;
2158
2159                 var_name = btf__name_by_offset(btf, t->name_off);
2160                 if (strcmp(var_name, ext_name))
2161                         continue;
2162
2163                 if (btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
2164                         return -EINVAL;
2165
2166                 return i;
2167         }
2168
2169         return -ENOENT;
2170 }
2171
2172 static enum extern_type find_extern_type(const struct btf *btf, int id,
2173                                          bool *is_signed)
2174 {
2175         const struct btf_type *t;
2176         const char *name;
2177
2178         t = skip_mods_and_typedefs(btf, id, NULL);
2179         name = btf__name_by_offset(btf, t->name_off);
2180
2181         if (is_signed)
2182                 *is_signed = false;
2183         switch (btf_kind(t)) {
2184         case BTF_KIND_INT: {
2185                 int enc = btf_int_encoding(t);
2186
2187                 if (enc & BTF_INT_BOOL)
2188                         return t->size == 1 ? EXT_BOOL : EXT_UNKNOWN;
2189                 if (is_signed)
2190                         *is_signed = enc & BTF_INT_SIGNED;
2191                 if (t->size == 1)
2192                         return EXT_CHAR;
2193                 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
2194                         return EXT_UNKNOWN;
2195                 return EXT_INT;
2196         }
2197         case BTF_KIND_ENUM:
2198                 if (t->size != 4)
2199                         return EXT_UNKNOWN;
2200                 if (strcmp(name, "libbpf_tristate"))
2201                         return EXT_UNKNOWN;
2202                 return EXT_TRISTATE;
2203         case BTF_KIND_ARRAY:
2204                 if (btf_array(t)->nelems == 0)
2205                         return EXT_UNKNOWN;
2206                 if (find_extern_type(btf, btf_array(t)->type, NULL) != EXT_CHAR)
2207                         return EXT_UNKNOWN;
2208                 return EXT_CHAR_ARR;
2209         default:
2210                 return EXT_UNKNOWN;
2211         }
2212 }
2213
2214 static int cmp_externs(const void *_a, const void *_b)
2215 {
2216         const struct extern_desc *a = _a;
2217         const struct extern_desc *b = _b;
2218
2219         /* descending order by alignment requirements */
2220         if (a->align != b->align)
2221                 return a->align > b->align ? -1 : 1;
2222         /* ascending order by size, within same alignment class */
2223         if (a->sz != b->sz)
2224                 return a->sz < b->sz ? -1 : 1;
2225         /* resolve ties by name */
2226         return strcmp(a->name, b->name);
2227 }
2228
2229 static int bpf_object__collect_externs(struct bpf_object *obj)
2230 {
2231         const struct btf_type *t;
2232         struct extern_desc *ext;
2233         int i, n, off, btf_id;
2234         struct btf_type *sec;
2235         const char *ext_name;
2236         Elf_Scn *scn;
2237         GElf_Shdr sh;
2238
2239         if (!obj->efile.symbols)
2240                 return 0;
2241
2242         scn = elf_getscn(obj->efile.elf, obj->efile.symbols_shndx);
2243         if (!scn)
2244                 return -LIBBPF_ERRNO__FORMAT;
2245         if (gelf_getshdr(scn, &sh) != &sh)
2246                 return -LIBBPF_ERRNO__FORMAT;
2247         n = sh.sh_size / sh.sh_entsize;
2248
2249         pr_debug("looking for externs among %d symbols...\n", n);
2250         for (i = 0; i < n; i++) {
2251                 GElf_Sym sym;
2252
2253                 if (!gelf_getsym(obj->efile.symbols, i, &sym))
2254                         return -LIBBPF_ERRNO__FORMAT;
2255                 if (!sym_is_extern(&sym))
2256                         continue;
2257                 ext_name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
2258                                       sym.st_name);
2259                 if (!ext_name || !ext_name[0])
2260                         continue;
2261
2262                 ext = obj->externs;
2263                 ext = reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
2264                 if (!ext)
2265                         return -ENOMEM;
2266                 obj->externs = ext;
2267                 ext = &ext[obj->nr_extern];
2268                 memset(ext, 0, sizeof(*ext));
2269                 obj->nr_extern++;
2270
2271                 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
2272                 if (ext->btf_id <= 0) {
2273                         pr_warn("failed to find BTF for extern '%s': %d\n",
2274                                 ext_name, ext->btf_id);
2275                         return ext->btf_id;
2276                 }
2277                 t = btf__type_by_id(obj->btf, ext->btf_id);
2278                 ext->name = btf__name_by_offset(obj->btf, t->name_off);
2279                 ext->sym_idx = i;
2280                 ext->is_weak = GELF_ST_BIND(sym.st_info) == STB_WEAK;
2281                 ext->sz = btf__resolve_size(obj->btf, t->type);
2282                 if (ext->sz <= 0) {
2283                         pr_warn("failed to resolve size of extern '%s': %d\n",
2284                                 ext_name, ext->sz);
2285                         return ext->sz;
2286                 }
2287                 ext->align = btf__align_of(obj->btf, t->type);
2288                 if (ext->align <= 0) {
2289                         pr_warn("failed to determine alignment of extern '%s': %d\n",
2290                                 ext_name, ext->align);
2291                         return -EINVAL;
2292                 }
2293                 ext->type = find_extern_type(obj->btf, t->type,
2294                                              &ext->is_signed);
2295                 if (ext->type == EXT_UNKNOWN) {
2296                         pr_warn("extern '%s' type is unsupported\n", ext_name);
2297                         return -ENOTSUP;
2298                 }
2299         }
2300         pr_debug("collected %d externs total\n", obj->nr_extern);
2301
2302         if (!obj->nr_extern)
2303                 return 0;
2304
2305         /* sort externs by (alignment, size, name) and calculate their offsets
2306          * within a map */
2307         qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
2308         off = 0;
2309         for (i = 0; i < obj->nr_extern; i++) {
2310                 ext = &obj->externs[i];
2311                 ext->data_off = roundup(off, ext->align);
2312                 off = ext->data_off + ext->sz;
2313                 pr_debug("extern #%d: symbol %d, off %u, name %s\n",
2314                          i, ext->sym_idx, ext->data_off, ext->name);
2315         }
2316
2317         btf_id = btf__find_by_name(obj->btf, KCONFIG_SEC);
2318         if (btf_id <= 0) {
2319                 pr_warn("no BTF info found for '%s' datasec\n", KCONFIG_SEC);
2320                 return -ESRCH;
2321         }
2322
2323         sec = (struct btf_type *)btf__type_by_id(obj->btf, btf_id);
2324         sec->size = off;
2325         n = btf_vlen(sec);
2326         for (i = 0; i < n; i++) {
2327                 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
2328
2329                 t = btf__type_by_id(obj->btf, vs->type);
2330                 ext_name = btf__name_by_offset(obj->btf, t->name_off);
2331                 ext = find_extern_by_name(obj, ext_name);
2332                 if (!ext) {
2333                         pr_warn("failed to find extern definition for BTF var '%s'\n",
2334                                 ext_name);
2335                         return -ESRCH;
2336                 }
2337                 vs->offset = ext->data_off;
2338                 btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
2339         }
2340
2341         return 0;
2342 }
2343
2344 static struct bpf_program *
2345 bpf_object__find_prog_by_idx(struct bpf_object *obj, int idx)
2346 {
2347         struct bpf_program *prog;
2348         size_t i;
2349
2350         for (i = 0; i < obj->nr_programs; i++) {
2351                 prog = &obj->programs[i];
2352                 if (prog->idx == idx)
2353                         return prog;
2354         }
2355         return NULL;
2356 }
2357
2358 struct bpf_program *
2359 bpf_object__find_program_by_title(const struct bpf_object *obj,
2360                                   const char *title)
2361 {
2362         struct bpf_program *pos;
2363
2364         bpf_object__for_each_program(pos, obj) {
2365                 if (pos->section_name && !strcmp(pos->section_name, title))
2366                         return pos;
2367         }
2368         return NULL;
2369 }
2370
2371 struct bpf_program *
2372 bpf_object__find_program_by_name(const struct bpf_object *obj,
2373                                  const char *name)
2374 {
2375         struct bpf_program *prog;
2376
2377         bpf_object__for_each_program(prog, obj) {
2378                 if (!strcmp(prog->name, name))
2379                         return prog;
2380         }
2381         return NULL;
2382 }
2383
2384 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
2385                                       int shndx)
2386 {
2387         return shndx == obj->efile.data_shndx ||
2388                shndx == obj->efile.bss_shndx ||
2389                shndx == obj->efile.rodata_shndx;
2390 }
2391
2392 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
2393                                       int shndx)
2394 {
2395         return shndx == obj->efile.maps_shndx ||
2396                shndx == obj->efile.btf_maps_shndx;
2397 }
2398
2399 static enum libbpf_map_type
2400 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
2401 {
2402         if (shndx == obj->efile.data_shndx)
2403                 return LIBBPF_MAP_DATA;
2404         else if (shndx == obj->efile.bss_shndx)
2405                 return LIBBPF_MAP_BSS;
2406         else if (shndx == obj->efile.rodata_shndx)
2407                 return LIBBPF_MAP_RODATA;
2408         else if (shndx == obj->efile.symbols_shndx)
2409                 return LIBBPF_MAP_KCONFIG;
2410         else
2411                 return LIBBPF_MAP_UNSPEC;
2412 }
2413
2414 static int bpf_program__record_reloc(struct bpf_program *prog,
2415                                      struct reloc_desc *reloc_desc,
2416                                      __u32 insn_idx, const char *name,
2417                                      const GElf_Sym *sym, const GElf_Rel *rel)
2418 {
2419         struct bpf_insn *insn = &prog->insns[insn_idx];
2420         size_t map_idx, nr_maps = prog->obj->nr_maps;
2421         struct bpf_object *obj = prog->obj;
2422         __u32 shdr_idx = sym->st_shndx;
2423         enum libbpf_map_type type;
2424         struct bpf_map *map;
2425
2426         /* sub-program call relocation */
2427         if (insn->code == (BPF_JMP | BPF_CALL)) {
2428                 if (insn->src_reg != BPF_PSEUDO_CALL) {
2429                         pr_warn("incorrect bpf_call opcode\n");
2430                         return -LIBBPF_ERRNO__RELOC;
2431                 }
2432                 /* text_shndx can be 0, if no default "main" program exists */
2433                 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
2434                         pr_warn("bad call relo against section %u\n", shdr_idx);
2435                         return -LIBBPF_ERRNO__RELOC;
2436                 }
2437                 if (sym->st_value % 8) {
2438                         pr_warn("bad call relo offset: %zu\n",
2439                                 (size_t)sym->st_value);
2440                         return -LIBBPF_ERRNO__RELOC;
2441                 }
2442                 reloc_desc->type = RELO_CALL;
2443                 reloc_desc->insn_idx = insn_idx;
2444                 reloc_desc->sym_off = sym->st_value;
2445                 obj->has_pseudo_calls = true;
2446                 return 0;
2447         }
2448
2449         if (insn->code != (BPF_LD | BPF_IMM | BPF_DW)) {
2450                 pr_warn("invalid relo for insns[%d].code 0x%x\n",
2451                         insn_idx, insn->code);
2452                 return -LIBBPF_ERRNO__RELOC;
2453         }
2454
2455         if (sym_is_extern(sym)) {
2456                 int sym_idx = GELF_R_SYM(rel->r_info);
2457                 int i, n = obj->nr_extern;
2458                 struct extern_desc *ext;
2459
2460                 for (i = 0; i < n; i++) {
2461                         ext = &obj->externs[i];
2462                         if (ext->sym_idx == sym_idx)
2463                                 break;
2464                 }
2465                 if (i >= n) {
2466                         pr_warn("extern relo failed to find extern for sym %d\n",
2467                                 sym_idx);
2468                         return -LIBBPF_ERRNO__RELOC;
2469                 }
2470                 pr_debug("found extern #%d '%s' (sym %d, off %u) for insn %u\n",
2471                          i, ext->name, ext->sym_idx, ext->data_off, insn_idx);
2472                 reloc_desc->type = RELO_EXTERN;
2473                 reloc_desc->insn_idx = insn_idx;
2474                 reloc_desc->sym_off = ext->data_off;
2475                 return 0;
2476         }
2477
2478         if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
2479                 pr_warn("invalid relo for \'%s\' in special section 0x%x; forgot to initialize global var?..\n",
2480                         name, shdr_idx);
2481                 return -LIBBPF_ERRNO__RELOC;
2482         }
2483
2484         type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
2485
2486         /* generic map reference relocation */
2487         if (type == LIBBPF_MAP_UNSPEC) {
2488                 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
2489                         pr_warn("bad map relo against section %u\n",
2490                                 shdr_idx);
2491                         return -LIBBPF_ERRNO__RELOC;
2492                 }
2493                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
2494                         map = &obj->maps[map_idx];
2495                         if (map->libbpf_type != type ||
2496                             map->sec_idx != sym->st_shndx ||
2497                             map->sec_offset != sym->st_value)
2498                                 continue;
2499                         pr_debug("found map %zd (%s, sec %d, off %zu) for insn %u\n",
2500                                  map_idx, map->name, map->sec_idx,
2501                                  map->sec_offset, insn_idx);
2502                         break;
2503                 }
2504                 if (map_idx >= nr_maps) {
2505                         pr_warn("map relo failed to find map for sec %u, off %zu\n",
2506                                 shdr_idx, (size_t)sym->st_value);
2507                         return -LIBBPF_ERRNO__RELOC;
2508                 }
2509                 reloc_desc->type = RELO_LD64;
2510                 reloc_desc->insn_idx = insn_idx;
2511                 reloc_desc->map_idx = map_idx;
2512                 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
2513                 return 0;
2514         }
2515
2516         /* global data map relocation */
2517         if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
2518                 pr_warn("bad data relo against section %u\n", shdr_idx);
2519                 return -LIBBPF_ERRNO__RELOC;
2520         }
2521         for (map_idx = 0; map_idx < nr_maps; map_idx++) {
2522                 map = &obj->maps[map_idx];
2523                 if (map->libbpf_type != type)
2524                         continue;
2525                 pr_debug("found data map %zd (%s, sec %d, off %zu) for insn %u\n",
2526                          map_idx, map->name, map->sec_idx, map->sec_offset,
2527                          insn_idx);
2528                 break;
2529         }
2530         if (map_idx >= nr_maps) {
2531                 pr_warn("data relo failed to find map for sec %u\n",
2532                         shdr_idx);
2533                 return -LIBBPF_ERRNO__RELOC;
2534         }
2535
2536         reloc_desc->type = RELO_DATA;
2537         reloc_desc->insn_idx = insn_idx;
2538         reloc_desc->map_idx = map_idx;
2539         reloc_desc->sym_off = sym->st_value;
2540         return 0;
2541 }
2542
2543 static int
2544 bpf_program__collect_reloc(struct bpf_program *prog, GElf_Shdr *shdr,
2545                            Elf_Data *data, struct bpf_object *obj)
2546 {
2547         Elf_Data *symbols = obj->efile.symbols;
2548         int err, i, nrels;
2549
2550         pr_debug("collecting relocating info for: '%s'\n", prog->section_name);
2551         nrels = shdr->sh_size / shdr->sh_entsize;
2552
2553         prog->reloc_desc = malloc(sizeof(*prog->reloc_desc) * nrels);
2554         if (!prog->reloc_desc) {
2555                 pr_warn("failed to alloc memory in relocation\n");
2556                 return -ENOMEM;
2557         }
2558         prog->nr_reloc = nrels;
2559
2560         for (i = 0; i < nrels; i++) {
2561                 const char *name;
2562                 __u32 insn_idx;
2563                 GElf_Sym sym;
2564                 GElf_Rel rel;
2565
2566                 if (!gelf_getrel(data, i, &rel)) {
2567                         pr_warn("relocation: failed to get %d reloc\n", i);
2568                         return -LIBBPF_ERRNO__FORMAT;
2569                 }
2570                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
2571                         pr_warn("relocation: symbol %"PRIx64" not found\n",
2572                                 GELF_R_SYM(rel.r_info));
2573                         return -LIBBPF_ERRNO__FORMAT;
2574                 }
2575                 if (rel.r_offset % sizeof(struct bpf_insn))
2576                         return -LIBBPF_ERRNO__FORMAT;
2577
2578                 insn_idx = rel.r_offset / sizeof(struct bpf_insn);
2579                 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx,
2580                                   sym.st_name) ? : "<?>";
2581
2582                 pr_debug("relo for shdr %u, symb %zu, value %zu, type %d, bind %d, name %d (\'%s\'), insn %u\n",
2583                          (__u32)sym.st_shndx, (size_t)GELF_R_SYM(rel.r_info),
2584                          (size_t)sym.st_value, GELF_ST_TYPE(sym.st_info),
2585                          GELF_ST_BIND(sym.st_info), sym.st_name, name,
2586                          insn_idx);
2587
2588                 err = bpf_program__record_reloc(prog, &prog->reloc_desc[i],
2589                                                 insn_idx, name, &sym, &rel);
2590                 if (err)
2591                         return err;
2592         }
2593         return 0;
2594 }
2595
2596 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
2597 {
2598         struct bpf_map_def *def = &map->def;
2599         __u32 key_type_id = 0, value_type_id = 0;
2600         int ret;
2601
2602         /* if it's BTF-defined map, we don't need to search for type IDs */
2603         if (map->sec_idx == obj->efile.btf_maps_shndx)
2604                 return 0;
2605
2606         if (!bpf_map__is_internal(map)) {
2607                 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
2608                                            def->value_size, &key_type_id,
2609                                            &value_type_id);
2610         } else {
2611                 /*
2612                  * LLVM annotates global data differently in BTF, that is,
2613                  * only as '.data', '.bss' or '.rodata'.
2614                  */
2615                 ret = btf__find_by_name(obj->btf,
2616                                 libbpf_type_to_btf_name[map->libbpf_type]);
2617         }
2618         if (ret < 0)
2619                 return ret;
2620
2621         map->btf_key_type_id = key_type_id;
2622         map->btf_value_type_id = bpf_map__is_internal(map) ?
2623                                  ret : value_type_id;
2624         return 0;
2625 }
2626
2627 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
2628 {
2629         struct bpf_map_info info = {};
2630         __u32 len = sizeof(info);
2631         int new_fd, err;
2632         char *new_name;
2633
2634         err = bpf_obj_get_info_by_fd(fd, &info, &len);
2635         if (err)
2636                 return err;
2637
2638         new_name = strdup(info.name);
2639         if (!new_name)
2640                 return -errno;
2641
2642         new_fd = open("/", O_RDONLY | O_CLOEXEC);
2643         if (new_fd < 0) {
2644                 err = -errno;
2645                 goto err_free_new_name;
2646         }
2647
2648         new_fd = dup3(fd, new_fd, O_CLOEXEC);
2649         if (new_fd < 0) {
2650                 err = -errno;
2651                 goto err_close_new_fd;
2652         }
2653
2654         err = zclose(map->fd);
2655         if (err) {
2656                 err = -errno;
2657                 goto err_close_new_fd;
2658         }
2659         free(map->name);
2660
2661         map->fd = new_fd;
2662         map->name = new_name;
2663         map->def.type = info.type;
2664         map->def.key_size = info.key_size;
2665         map->def.value_size = info.value_size;
2666         map->def.max_entries = info.max_entries;
2667         map->def.map_flags = info.map_flags;
2668         map->btf_key_type_id = info.btf_key_type_id;
2669         map->btf_value_type_id = info.btf_value_type_id;
2670         map->reused = true;
2671
2672         return 0;
2673
2674 err_close_new_fd:
2675         close(new_fd);
2676 err_free_new_name:
2677         free(new_name);
2678         return err;
2679 }
2680
2681 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
2682 {
2683         if (!map || !max_entries)
2684                 return -EINVAL;
2685
2686         /* If map already created, its attributes can't be changed. */
2687         if (map->fd >= 0)
2688                 return -EBUSY;
2689
2690         map->def.max_entries = max_entries;
2691
2692         return 0;
2693 }
2694
2695 static int
2696 bpf_object__probe_name(struct bpf_object *obj)
2697 {
2698         struct bpf_load_program_attr attr;
2699         char *cp, errmsg[STRERR_BUFSIZE];
2700         struct bpf_insn insns[] = {
2701                 BPF_MOV64_IMM(BPF_REG_0, 0),
2702                 BPF_EXIT_INSN(),
2703         };
2704         int ret;
2705
2706         /* make sure basic loading works */
2707
2708         memset(&attr, 0, sizeof(attr));
2709         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
2710         attr.insns = insns;
2711         attr.insns_cnt = ARRAY_SIZE(insns);
2712         attr.license = "GPL";
2713
2714         ret = bpf_load_program_xattr(&attr, NULL, 0);
2715         if (ret < 0) {
2716                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2717                 pr_warn("Error in %s():%s(%d). Couldn't load basic 'r0 = 0' BPF program.\n",
2718                         __func__, cp, errno);
2719                 return -errno;
2720         }
2721         close(ret);
2722
2723         /* now try the same program, but with the name */
2724
2725         attr.name = "test";
2726         ret = bpf_load_program_xattr(&attr, NULL, 0);
2727         if (ret >= 0) {
2728                 obj->caps.name = 1;
2729                 close(ret);
2730         }
2731
2732         return 0;
2733 }
2734
2735 static int
2736 bpf_object__probe_global_data(struct bpf_object *obj)
2737 {
2738         struct bpf_load_program_attr prg_attr;
2739         struct bpf_create_map_attr map_attr;
2740         char *cp, errmsg[STRERR_BUFSIZE];
2741         struct bpf_insn insns[] = {
2742                 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
2743                 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
2744                 BPF_MOV64_IMM(BPF_REG_0, 0),
2745                 BPF_EXIT_INSN(),
2746         };
2747         int ret, map;
2748
2749         memset(&map_attr, 0, sizeof(map_attr));
2750         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
2751         map_attr.key_size = sizeof(int);
2752         map_attr.value_size = 32;
2753         map_attr.max_entries = 1;
2754
2755         map = bpf_create_map_xattr(&map_attr);
2756         if (map < 0) {
2757                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
2758                 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
2759                         __func__, cp, errno);
2760                 return -errno;
2761         }
2762
2763         insns[0].imm = map;
2764
2765         memset(&prg_attr, 0, sizeof(prg_attr));
2766         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
2767         prg_attr.insns = insns;
2768         prg_attr.insns_cnt = ARRAY_SIZE(insns);
2769         prg_attr.license = "GPL";
2770
2771         ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
2772         if (ret >= 0) {
2773                 obj->caps.global_data = 1;
2774                 close(ret);
2775         }
2776
2777         close(map);
2778         return 0;
2779 }
2780
2781 static int bpf_object__probe_btf_func(struct bpf_object *obj)
2782 {
2783         static const char strs[] = "\0int\0x\0a";
2784         /* void x(int a) {} */
2785         __u32 types[] = {
2786                 /* int */
2787                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
2788                 /* FUNC_PROTO */                                /* [2] */
2789                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
2790                 BTF_PARAM_ENC(7, 1),
2791                 /* FUNC x */                                    /* [3] */
2792                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
2793         };
2794         int btf_fd;
2795
2796         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2797                                       strs, sizeof(strs));
2798         if (btf_fd >= 0) {
2799                 obj->caps.btf_func = 1;
2800                 close(btf_fd);
2801                 return 1;
2802         }
2803
2804         return 0;
2805 }
2806
2807 static int bpf_object__probe_btf_datasec(struct bpf_object *obj)
2808 {
2809         static const char strs[] = "\0x\0.data";
2810         /* static int a; */
2811         __u32 types[] = {
2812                 /* int */
2813                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
2814                 /* VAR x */                                     /* [2] */
2815                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
2816                 BTF_VAR_STATIC,
2817                 /* DATASEC val */                               /* [3] */
2818                 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
2819                 BTF_VAR_SECINFO_ENC(2, 0, 4),
2820         };
2821         int btf_fd;
2822
2823         btf_fd = libbpf__load_raw_btf((char *)types, sizeof(types),
2824                                       strs, sizeof(strs));
2825         if (btf_fd >= 0) {
2826                 obj->caps.btf_datasec = 1;
2827                 close(btf_fd);
2828                 return 1;
2829         }
2830
2831         return 0;
2832 }
2833
2834 static int bpf_object__probe_array_mmap(struct bpf_object *obj)
2835 {
2836         struct bpf_create_map_attr attr = {
2837                 .map_type = BPF_MAP_TYPE_ARRAY,
2838                 .map_flags = BPF_F_MMAPABLE,
2839                 .key_size = sizeof(int),
2840                 .value_size = sizeof(int),
2841                 .max_entries = 1,
2842         };
2843         int fd;
2844
2845         fd = bpf_create_map_xattr(&attr);
2846         if (fd >= 0) {
2847                 obj->caps.array_mmap = 1;
2848                 close(fd);
2849                 return 1;
2850         }
2851
2852         return 0;
2853 }
2854
2855 static int
2856 bpf_object__probe_caps(struct bpf_object *obj)
2857 {
2858         int (*probe_fn[])(struct bpf_object *obj) = {
2859                 bpf_object__probe_name,
2860                 bpf_object__probe_global_data,
2861                 bpf_object__probe_btf_func,
2862                 bpf_object__probe_btf_datasec,
2863                 bpf_object__probe_array_mmap,
2864         };
2865         int i, ret;
2866
2867         for (i = 0; i < ARRAY_SIZE(probe_fn); i++) {
2868                 ret = probe_fn[i](obj);
2869                 if (ret < 0)
2870                         pr_debug("Probe #%d failed with %d.\n", i, ret);
2871         }
2872
2873         return 0;
2874 }
2875
2876 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
2877 {
2878         struct bpf_map_info map_info = {};
2879         char msg[STRERR_BUFSIZE];
2880         __u32 map_info_len;
2881
2882         map_info_len = sizeof(map_info);
2883
2884         if (bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len)) {
2885                 pr_warn("failed to get map info for map FD %d: %s\n",
2886                         map_fd, libbpf_strerror_r(errno, msg, sizeof(msg)));
2887                 return false;
2888         }
2889
2890         return (map_info.type == map->def.type &&
2891                 map_info.key_size == map->def.key_size &&
2892                 map_info.value_size == map->def.value_size &&
2893                 map_info.max_entries == map->def.max_entries &&
2894                 map_info.map_flags == map->def.map_flags);
2895 }
2896
2897 static int
2898 bpf_object__reuse_map(struct bpf_map *map)
2899 {
2900         char *cp, errmsg[STRERR_BUFSIZE];
2901         int err, pin_fd;
2902
2903         pin_fd = bpf_obj_get(map->pin_path);
2904         if (pin_fd < 0) {
2905                 err = -errno;
2906                 if (err == -ENOENT) {
2907                         pr_debug("found no pinned map to reuse at '%s'\n",
2908                                  map->pin_path);
2909                         return 0;
2910                 }
2911
2912                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
2913                 pr_warn("couldn't retrieve pinned map '%s': %s\n",
2914                         map->pin_path, cp);
2915                 return err;
2916         }
2917
2918         if (!map_is_reuse_compat(map, pin_fd)) {
2919                 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
2920                         map->pin_path);
2921                 close(pin_fd);
2922                 return -EINVAL;
2923         }
2924
2925         err = bpf_map__reuse_fd(map, pin_fd);
2926         if (err) {
2927                 close(pin_fd);
2928                 return err;
2929         }
2930         map->pinned = true;
2931         pr_debug("reused pinned map at '%s'\n", map->pin_path);
2932
2933         return 0;
2934 }
2935
2936 static int
2937 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
2938 {
2939         enum libbpf_map_type map_type = map->libbpf_type;
2940         char *cp, errmsg[STRERR_BUFSIZE];
2941         int err, zero = 0;
2942
2943         /* kernel already zero-initializes .bss map. */
2944         if (map_type == LIBBPF_MAP_BSS)
2945                 return 0;
2946
2947         err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
2948         if (err) {
2949                 err = -errno;
2950                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2951                 pr_warn("Error setting initial map(%s) contents: %s\n",
2952                         map->name, cp);
2953                 return err;
2954         }
2955
2956         /* Freeze .rodata and .kconfig map as read-only from syscall side. */
2957         if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
2958                 err = bpf_map_freeze(map->fd);
2959                 if (err) {
2960                         err = -errno;
2961                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
2962                         pr_warn("Error freezing map(%s) as read-only: %s\n",
2963                                 map->name, cp);
2964                         return err;
2965                 }
2966         }
2967         return 0;
2968 }
2969
2970 static int
2971 bpf_object__create_maps(struct bpf_object *obj)
2972 {
2973         struct bpf_create_map_attr create_attr = {};
2974         int nr_cpus = 0;
2975         unsigned int i;
2976         int err;
2977
2978         for (i = 0; i < obj->nr_maps; i++) {
2979                 struct bpf_map *map = &obj->maps[i];
2980                 struct bpf_map_def *def = &map->def;
2981                 char *cp, errmsg[STRERR_BUFSIZE];
2982                 int *pfd = &map->fd;
2983
2984                 if (map->pin_path) {
2985                         err = bpf_object__reuse_map(map);
2986                         if (err) {
2987                                 pr_warn("error reusing pinned map %s\n",
2988                                         map->name);
2989                                 return err;
2990                         }
2991                 }
2992
2993                 if (map->fd >= 0) {
2994                         pr_debug("skip map create (preset) %s: fd=%d\n",
2995                                  map->name, map->fd);
2996                         continue;
2997                 }
2998
2999                 if (obj->caps.name)
3000                         create_attr.name = map->name;
3001                 create_attr.map_ifindex = map->map_ifindex;
3002                 create_attr.map_type = def->type;
3003                 create_attr.map_flags = def->map_flags;
3004                 create_attr.key_size = def->key_size;
3005                 create_attr.value_size = def->value_size;
3006                 if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY &&
3007                     !def->max_entries) {
3008                         if (!nr_cpus)
3009                                 nr_cpus = libbpf_num_possible_cpus();
3010                         if (nr_cpus < 0) {
3011                                 pr_warn("failed to determine number of system CPUs: %d\n",
3012                                         nr_cpus);
3013                                 err = nr_cpus;
3014                                 goto err_out;
3015                         }
3016                         pr_debug("map '%s': setting size to %d\n",
3017                                  map->name, nr_cpus);
3018                         create_attr.max_entries = nr_cpus;
3019                 } else {
3020                         create_attr.max_entries = def->max_entries;
3021                 }
3022                 create_attr.btf_fd = 0;
3023                 create_attr.btf_key_type_id = 0;
3024                 create_attr.btf_value_type_id = 0;
3025                 if (bpf_map_type__is_map_in_map(def->type) &&
3026                     map->inner_map_fd >= 0)
3027                         create_attr.inner_map_fd = map->inner_map_fd;
3028
3029                 if (obj->btf && !bpf_map_find_btf_info(obj, map)) {
3030                         create_attr.btf_fd = btf__fd(obj->btf);
3031                         create_attr.btf_key_type_id = map->btf_key_type_id;
3032                         create_attr.btf_value_type_id = map->btf_value_type_id;
3033                 }
3034
3035                 *pfd = bpf_create_map_xattr(&create_attr);
3036                 if (*pfd < 0 && (create_attr.btf_key_type_id ||
3037                                  create_attr.btf_value_type_id)) {
3038                         err = -errno;
3039                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3040                         pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
3041                                 map->name, cp, err);
3042                         create_attr.btf_fd = 0;
3043                         create_attr.btf_key_type_id = 0;
3044                         create_attr.btf_value_type_id = 0;
3045                         map->btf_key_type_id = 0;
3046                         map->btf_value_type_id = 0;
3047                         *pfd = bpf_create_map_xattr(&create_attr);
3048                 }
3049
3050                 if (*pfd < 0) {
3051                         size_t j;
3052
3053                         err = -errno;
3054 err_out:
3055                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
3056                         pr_warn("failed to create map (name: '%s'): %s(%d)\n",
3057                                 map->name, cp, err);
3058                         pr_perm_msg(err);
3059                         for (j = 0; j < i; j++)
3060                                 zclose(obj->maps[j].fd);
3061                         return err;
3062                 }
3063
3064                 if (bpf_map__is_internal(map)) {
3065                         err = bpf_object__populate_internal_map(obj, map);
3066                         if (err < 0) {
3067                                 zclose(*pfd);
3068                                 goto err_out;
3069                         }
3070                 }
3071
3072                 if (map->pin_path && !map->pinned) {
3073                         err = bpf_map__pin(map, NULL);
3074                         if (err) {
3075                                 pr_warn("failed to auto-pin map name '%s' at '%s'\n",
3076                                         map->name, map->pin_path);
3077                                 return err;
3078                         }
3079                 }
3080
3081                 pr_debug("created map %s: fd=%d\n", map->name, *pfd);
3082         }
3083
3084         return 0;
3085 }
3086
3087 static int
3088 check_btf_ext_reloc_err(struct bpf_program *prog, int err,
3089                         void *btf_prog_info, const char *info_name)
3090 {
3091         if (err != -ENOENT) {
3092                 pr_warn("Error in loading %s for sec %s.\n",
3093                         info_name, prog->section_name);
3094                 return err;
3095         }
3096
3097         /* err == -ENOENT (i.e. prog->section_name not found in btf_ext) */
3098
3099         if (btf_prog_info) {
3100                 /*
3101                  * Some info has already been found but has problem
3102                  * in the last btf_ext reloc. Must have to error out.
3103                  */
3104                 pr_warn("Error in relocating %s for sec %s.\n",
3105                         info_name, prog->section_name);
3106                 return err;
3107         }
3108
3109         /* Have problem loading the very first info. Ignore the rest. */
3110         pr_warn("Cannot find %s for main program sec %s. Ignore all %s.\n",
3111                 info_name, prog->section_name, info_name);
3112         return 0;
3113 }
3114
3115 static int
3116 bpf_program_reloc_btf_ext(struct bpf_program *prog, struct bpf_object *obj,
3117                           const char *section_name,  __u32 insn_offset)
3118 {
3119         int err;
3120
3121         if (!insn_offset || prog->func_info) {
3122                 /*
3123                  * !insn_offset => main program
3124                  *
3125                  * For sub prog, the main program's func_info has to
3126                  * be loaded first (i.e. prog->func_info != NULL)
3127                  */
3128                 err = btf_ext__reloc_func_info(obj->btf, obj->btf_ext,
3129                                                section_name, insn_offset,
3130                                                &prog->func_info,
3131                                                &prog->func_info_cnt);
3132                 if (err)
3133                         return check_btf_ext_reloc_err(prog, err,
3134                                                        prog->func_info,
3135                                                        "bpf_func_info");
3136
3137                 prog->func_info_rec_size = btf_ext__func_info_rec_size(obj->btf_ext);
3138         }
3139
3140         if (!insn_offset || prog->line_info) {
3141                 err = btf_ext__reloc_line_info(obj->btf, obj->btf_ext,
3142                                                section_name, insn_offset,
3143                                                &prog->line_info,
3144                                                &prog->line_info_cnt);
3145                 if (err)
3146                         return check_btf_ext_reloc_err(prog, err,
3147                                                        prog->line_info,
3148                                                        "bpf_line_info");
3149
3150                 prog->line_info_rec_size = btf_ext__line_info_rec_size(obj->btf_ext);
3151         }
3152
3153         return 0;
3154 }
3155
3156 #define BPF_CORE_SPEC_MAX_LEN 64
3157
3158 /* represents BPF CO-RE field or array element accessor */
3159 struct bpf_core_accessor {
3160         __u32 type_id;          /* struct/union type or array element type */
3161         __u32 idx;              /* field index or array index */
3162         const char *name;       /* field name or NULL for array accessor */
3163 };
3164
3165 struct bpf_core_spec {
3166         const struct btf *btf;
3167         /* high-level spec: named fields and array indices only */
3168         struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
3169         /* high-level spec length */
3170         int len;
3171         /* raw, low-level spec: 1-to-1 with accessor spec string */
3172         int raw_spec[BPF_CORE_SPEC_MAX_LEN];
3173         /* raw spec length */
3174         int raw_len;
3175         /* field bit offset represented by spec */
3176         __u32 bit_offset;
3177 };
3178
3179 static bool str_is_empty(const char *s)
3180 {
3181         return !s || !s[0];
3182 }
3183
3184 static bool is_flex_arr(const struct btf *btf,
3185                         const struct bpf_core_accessor *acc,
3186                         const struct btf_array *arr)
3187 {
3188         const struct btf_type *t;
3189
3190         /* not a flexible array, if not inside a struct or has non-zero size */
3191         if (!acc->name || arr->nelems > 0)
3192                 return false;
3193
3194         /* has to be the last member of enclosing struct */
3195         t = btf__type_by_id(btf, acc->type_id);
3196         return acc->idx == btf_vlen(t) - 1;
3197 }
3198
3199 /*
3200  * Turn bpf_field_reloc into a low- and high-level spec representation,
3201  * validating correctness along the way, as well as calculating resulting
3202  * field bit offset, specified by accessor string. Low-level spec captures
3203  * every single level of nestedness, including traversing anonymous
3204  * struct/union members. High-level one only captures semantically meaningful
3205  * "turning points": named fields and array indicies.
3206  * E.g., for this case:
3207  *
3208  *   struct sample {
3209  *       int __unimportant;
3210  *       struct {
3211  *           int __1;
3212  *           int __2;
3213  *           int a[7];
3214  *       };
3215  *   };
3216  *
3217  *   struct sample *s = ...;
3218  *
3219  *   int x = &s->a[3]; // access string = '0:1:2:3'
3220  *
3221  * Low-level spec has 1:1 mapping with each element of access string (it's
3222  * just a parsed access string representation): [0, 1, 2, 3].
3223  *
3224  * High-level spec will capture only 3 points:
3225  *   - intial zero-index access by pointer (&s->... is the same as &s[0]...);
3226  *   - field 'a' access (corresponds to '2' in low-level spec);
3227  *   - array element #3 access (corresponds to '3' in low-level spec).
3228  *
3229  */
3230 static int bpf_core_spec_parse(const struct btf *btf,
3231                                __u32 type_id,
3232                                const char *spec_str,
3233                                struct bpf_core_spec *spec)
3234 {
3235         int access_idx, parsed_len, i;
3236         struct bpf_core_accessor *acc;
3237         const struct btf_type *t;
3238         const char *name;
3239         __u32 id;
3240         __s64 sz;
3241
3242         if (str_is_empty(spec_str) || *spec_str == ':')
3243                 return -EINVAL;
3244
3245         memset(spec, 0, sizeof(*spec));
3246         spec->btf = btf;
3247
3248         /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
3249         while (*spec_str) {
3250                 if (*spec_str == ':')
3251                         ++spec_str;
3252                 if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
3253                         return -EINVAL;
3254                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
3255                         return -E2BIG;
3256                 spec_str += parsed_len;
3257                 spec->raw_spec[spec->raw_len++] = access_idx;
3258         }
3259
3260         if (spec->raw_len == 0)
3261                 return -EINVAL;
3262
3263         /* first spec value is always reloc type array index */
3264         t = skip_mods_and_typedefs(btf, type_id, &id);
3265         if (!t)
3266                 return -EINVAL;
3267
3268         access_idx = spec->raw_spec[0];
3269         spec->spec[0].type_id = id;
3270         spec->spec[0].idx = access_idx;
3271         spec->len++;
3272
3273         sz = btf__resolve_size(btf, id);
3274         if (sz < 0)
3275                 return sz;
3276         spec->bit_offset = access_idx * sz * 8;
3277
3278         for (i = 1; i < spec->raw_len; i++) {
3279                 t = skip_mods_and_typedefs(btf, id, &id);
3280                 if (!t)
3281                         return -EINVAL;
3282
3283                 access_idx = spec->raw_spec[i];
3284                 acc = &spec->spec[spec->len];
3285
3286                 if (btf_is_composite(t)) {
3287                         const struct btf_member *m;
3288                         __u32 bit_offset;
3289
3290                         if (access_idx >= btf_vlen(t))
3291                                 return -EINVAL;
3292
3293                         bit_offset = btf_member_bit_offset(t, access_idx);
3294                         spec->bit_offset += bit_offset;
3295
3296                         m = btf_members(t) + access_idx;
3297                         if (m->name_off) {
3298                                 name = btf__name_by_offset(btf, m->name_off);
3299                                 if (str_is_empty(name))
3300                                         return -EINVAL;
3301
3302                                 acc->type_id = id;
3303                                 acc->idx = access_idx;
3304                                 acc->name = name;
3305                                 spec->len++;
3306                         }
3307
3308                         id = m->type;
3309                 } else if (btf_is_array(t)) {
3310                         const struct btf_array *a = btf_array(t);
3311                         bool flex;
3312
3313                         t = skip_mods_and_typedefs(btf, a->type, &id);
3314                         if (!t)
3315                                 return -EINVAL;
3316
3317                         flex = is_flex_arr(btf, acc - 1, a);
3318                         if (!flex && access_idx >= a->nelems)
3319                                 return -EINVAL;
3320
3321                         spec->spec[spec->len].type_id = id;
3322                         spec->spec[spec->len].idx = access_idx;
3323                         spec->len++;
3324
3325                         sz = btf__resolve_size(btf, id);
3326                         if (sz < 0)
3327                                 return sz;
3328                         spec->bit_offset += access_idx * sz * 8;
3329                 } else {
3330                         pr_warn("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %d\n",
3331                                 type_id, spec_str, i, id, btf_kind(t));
3332                         return -EINVAL;
3333                 }
3334         }
3335
3336         return 0;
3337 }
3338
3339 static bool bpf_core_is_flavor_sep(const char *s)
3340 {
3341         /* check X___Y name pattern, where X and Y are not underscores */
3342         return s[0] != '_' &&                                 /* X */
3343                s[1] == '_' && s[2] == '_' && s[3] == '_' &&   /* ___ */
3344                s[4] != '_';                                   /* Y */
3345 }
3346
3347 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
3348  * before last triple underscore. Struct name part after last triple
3349  * underscore is ignored by BPF CO-RE relocation during relocation matching.
3350  */
3351 static size_t bpf_core_essential_name_len(const char *name)
3352 {
3353         size_t n = strlen(name);
3354         int i;
3355
3356         for (i = n - 5; i >= 0; i--) {
3357                 if (bpf_core_is_flavor_sep(name + i))
3358                         return i + 1;
3359         }
3360         return n;
3361 }
3362
3363 /* dynamically sized list of type IDs */
3364 struct ids_vec {
3365         __u32 *data;
3366         int len;
3367 };
3368
3369 static void bpf_core_free_cands(struct ids_vec *cand_ids)
3370 {
3371         free(cand_ids->data);
3372         free(cand_ids);
3373 }
3374
3375 static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
3376                                            __u32 local_type_id,
3377                                            const struct btf *targ_btf)
3378 {
3379         size_t local_essent_len, targ_essent_len;
3380         const char *local_name, *targ_name;
3381         const struct btf_type *t;
3382         struct ids_vec *cand_ids;
3383         __u32 *new_ids;
3384         int i, err, n;
3385
3386         t = btf__type_by_id(local_btf, local_type_id);
3387         if (!t)
3388                 return ERR_PTR(-EINVAL);
3389
3390         local_name = btf__name_by_offset(local_btf, t->name_off);
3391         if (str_is_empty(local_name))
3392                 return ERR_PTR(-EINVAL);
3393         local_essent_len = bpf_core_essential_name_len(local_name);
3394
3395         cand_ids = calloc(1, sizeof(*cand_ids));
3396         if (!cand_ids)
3397                 return ERR_PTR(-ENOMEM);
3398
3399         n = btf__get_nr_types(targ_btf);
3400         for (i = 1; i <= n; i++) {
3401                 t = btf__type_by_id(targ_btf, i);
3402                 targ_name = btf__name_by_offset(targ_btf, t->name_off);
3403                 if (str_is_empty(targ_name))
3404                         continue;
3405
3406                 targ_essent_len = bpf_core_essential_name_len(targ_name);
3407                 if (targ_essent_len != local_essent_len)
3408                         continue;
3409
3410                 if (strncmp(local_name, targ_name, local_essent_len) == 0) {
3411                         pr_debug("[%d] %s: found candidate [%d] %s\n",
3412                                  local_type_id, local_name, i, targ_name);
3413                         new_ids = realloc(cand_ids->data, cand_ids->len + 1);
3414                         if (!new_ids) {
3415                                 err = -ENOMEM;
3416                                 goto err_out;
3417                         }
3418                         cand_ids->data = new_ids;
3419                         cand_ids->data[cand_ids->len++] = i;
3420                 }
3421         }
3422         return cand_ids;
3423 err_out:
3424         bpf_core_free_cands(cand_ids);
3425         return ERR_PTR(err);
3426 }
3427
3428 /* Check two types for compatibility, skipping const/volatile/restrict and
3429  * typedefs, to ensure we are relocating compatible entities:
3430  *   - any two STRUCTs/UNIONs are compatible and can be mixed;
3431  *   - any two FWDs are compatible, if their names match (modulo flavor suffix);
3432  *   - any two PTRs are always compatible;
3433  *   - for ENUMs, names should be the same (ignoring flavor suffix) or at
3434  *     least one of enums should be anonymous;
3435  *   - for ENUMs, check sizes, names are ignored;
3436  *   - for INT, size and signedness are ignored;
3437  *   - for ARRAY, dimensionality is ignored, element types are checked for
3438  *     compatibility recursively;
3439  *   - everything else shouldn't be ever a target of relocation.
3440  * These rules are not set in stone and probably will be adjusted as we get
3441  * more experience with using BPF CO-RE relocations.
3442  */
3443 static int bpf_core_fields_are_compat(const struct btf *local_btf,
3444                                       __u32 local_id,
3445                                       const struct btf *targ_btf,
3446                                       __u32 targ_id)
3447 {
3448         const struct btf_type *local_type, *targ_type;
3449
3450 recur:
3451         local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
3452         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
3453         if (!local_type || !targ_type)
3454                 return -EINVAL;
3455
3456         if (btf_is_composite(local_type) && btf_is_composite(targ_type))
3457                 return 1;
3458         if (btf_kind(local_type) != btf_kind(targ_type))
3459                 return 0;
3460
3461         switch (btf_kind(local_type)) {
3462         case BTF_KIND_PTR:
3463                 return 1;
3464         case BTF_KIND_FWD:
3465         case BTF_KIND_ENUM: {
3466                 const char *local_name, *targ_name;
3467                 size_t local_len, targ_len;
3468
3469                 local_name = btf__name_by_offset(local_btf,
3470                                                  local_type->name_off);
3471                 targ_name = btf__name_by_offset(targ_btf, targ_type->name_off);
3472                 local_len = bpf_core_essential_name_len(local_name);
3473                 targ_len = bpf_core_essential_name_len(targ_name);
3474                 /* one of them is anonymous or both w/ same flavor-less names */
3475                 return local_len == 0 || targ_len == 0 ||
3476                        (local_len == targ_len &&
3477                         strncmp(local_name, targ_name, local_len) == 0);
3478         }
3479         case BTF_KIND_INT:
3480                 /* just reject deprecated bitfield-like integers; all other
3481                  * integers are by default compatible between each other
3482                  */
3483                 return btf_int_offset(local_type) == 0 &&
3484                        btf_int_offset(targ_type) == 0;
3485         case BTF_KIND_ARRAY:
3486                 local_id = btf_array(local_type)->type;
3487                 targ_id = btf_array(targ_type)->type;
3488                 goto recur;
3489         default:
3490                 pr_warn("unexpected kind %d relocated, local [%d], target [%d]\n",
3491                         btf_kind(local_type), local_id, targ_id);
3492                 return 0;
3493         }
3494 }
3495
3496 /*
3497  * Given single high-level named field accessor in local type, find
3498  * corresponding high-level accessor for a target type. Along the way,
3499  * maintain low-level spec for target as well. Also keep updating target
3500  * bit offset.
3501  *
3502  * Searching is performed through recursive exhaustive enumeration of all
3503  * fields of a struct/union. If there are any anonymous (embedded)
3504  * structs/unions, they are recursively searched as well. If field with
3505  * desired name is found, check compatibility between local and target types,
3506  * before returning result.
3507  *
3508  * 1 is returned, if field is found.
3509  * 0 is returned if no compatible field is found.
3510  * <0 is returned on error.
3511  */
3512 static int bpf_core_match_member(const struct btf *local_btf,
3513                                  const struct bpf_core_accessor *local_acc,
3514                                  const struct btf *targ_btf,
3515                                  __u32 targ_id,
3516                                  struct bpf_core_spec *spec,
3517                                  __u32 *next_targ_id)
3518 {
3519         const struct btf_type *local_type, *targ_type;
3520         const struct btf_member *local_member, *m;
3521         const char *local_name, *targ_name;
3522         __u32 local_id;
3523         int i, n, found;
3524
3525         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
3526         if (!targ_type)
3527                 return -EINVAL;
3528         if (!btf_is_composite(targ_type))
3529                 return 0;
3530
3531         local_id = local_acc->type_id;
3532         local_type = btf__type_by_id(local_btf, local_id);
3533         local_member = btf_members(local_type) + local_acc->idx;
3534         local_name = btf__name_by_offset(local_btf, local_member->name_off);
3535
3536         n = btf_vlen(targ_type);
3537         m = btf_members(targ_type);
3538         for (i = 0; i < n; i++, m++) {
3539                 __u32 bit_offset;
3540
3541                 bit_offset = btf_member_bit_offset(targ_type, i);
3542
3543                 /* too deep struct/union/array nesting */
3544                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
3545                         return -E2BIG;
3546
3547                 /* speculate this member will be the good one */
3548                 spec->bit_offset += bit_offset;
3549                 spec->raw_spec[spec->raw_len++] = i;
3550
3551                 targ_name = btf__name_by_offset(targ_btf, m->name_off);
3552                 if (str_is_empty(targ_name)) {
3553                         /* embedded struct/union, we need to go deeper */
3554                         found = bpf_core_match_member(local_btf, local_acc,
3555                                                       targ_btf, m->type,
3556                                                       spec, next_targ_id);
3557                         if (found) /* either found or error */
3558                                 return found;
3559                 } else if (strcmp(local_name, targ_name) == 0) {
3560                         /* matching named field */
3561                         struct bpf_core_accessor *targ_acc;
3562
3563                         targ_acc = &spec->spec[spec->len++];
3564                         targ_acc->type_id = targ_id;
3565                         targ_acc->idx = i;
3566                         targ_acc->name = targ_name;
3567
3568                         *next_targ_id = m->type;
3569                         found = bpf_core_fields_are_compat(local_btf,
3570                                                            local_member->type,
3571                                                            targ_btf, m->type);
3572                         if (!found)
3573                                 spec->len--; /* pop accessor */
3574                         return found;
3575                 }
3576                 /* member turned out not to be what we looked for */
3577                 spec->bit_offset -= bit_offset;
3578                 spec->raw_len--;
3579         }
3580
3581         return 0;
3582 }
3583
3584 /*
3585  * Try to match local spec to a target type and, if successful, produce full
3586  * target spec (high-level, low-level + bit offset).
3587  */
3588 static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
3589                                const struct btf *targ_btf, __u32 targ_id,
3590                                struct bpf_core_spec *targ_spec)
3591 {
3592         const struct btf_type *targ_type;
3593         const struct bpf_core_accessor *local_acc;
3594         struct bpf_core_accessor *targ_acc;
3595         int i, sz, matched;
3596
3597         memset(targ_spec, 0, sizeof(*targ_spec));
3598         targ_spec->btf = targ_btf;
3599
3600         local_acc = &local_spec->spec[0];
3601         targ_acc = &targ_spec->spec[0];
3602
3603         for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) {
3604                 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id,
3605                                                    &targ_id);
3606                 if (!targ_type)
3607                         return -EINVAL;
3608
3609                 if (local_acc->name) {
3610                         matched = bpf_core_match_member(local_spec->btf,
3611                                                         local_acc,
3612                                                         targ_btf, targ_id,
3613                                                         targ_spec, &targ_id);
3614                         if (matched <= 0)
3615                                 return matched;
3616                 } else {
3617                         /* for i=0, targ_id is already treated as array element
3618                          * type (because it's the original struct), for others
3619                          * we should find array element type first
3620                          */
3621                         if (i > 0) {
3622                                 const struct btf_array *a;
3623                                 bool flex;
3624
3625                                 if (!btf_is_array(targ_type))
3626                                         return 0;
3627
3628                                 a = btf_array(targ_type);
3629                                 flex = is_flex_arr(targ_btf, targ_acc - 1, a);
3630                                 if (!flex && local_acc->idx >= a->nelems)
3631                                         return 0;
3632                                 if (!skip_mods_and_typedefs(targ_btf, a->type,
3633                                                             &targ_id))
3634                                         return -EINVAL;
3635                         }
3636
3637                         /* too deep struct/union/array nesting */
3638                         if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
3639                                 return -E2BIG;
3640
3641                         targ_acc->type_id = targ_id;
3642                         targ_acc->idx = local_acc->idx;
3643                         targ_acc->name = NULL;
3644                         targ_spec->len++;
3645                         targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
3646                         targ_spec->raw_len++;
3647
3648                         sz = btf__resolve_size(targ_btf, targ_id);
3649                         if (sz < 0)
3650                                 return sz;
3651                         targ_spec->bit_offset += local_acc->idx * sz * 8;
3652                 }
3653         }
3654
3655         return 1;
3656 }
3657
3658 static int bpf_core_calc_field_relo(const struct bpf_program *prog,
3659                                     const struct bpf_field_reloc *relo,
3660                                     const struct bpf_core_spec *spec,
3661                                     __u32 *val, bool *validate)
3662 {
3663         const struct bpf_core_accessor *acc = &spec->spec[spec->len - 1];
3664         const struct btf_type *t = btf__type_by_id(spec->btf, acc->type_id);
3665         __u32 byte_off, byte_sz, bit_off, bit_sz;
3666         const struct btf_member *m;
3667         const struct btf_type *mt;
3668         bool bitfield;
3669         __s64 sz;
3670
3671         /* a[n] accessor needs special handling */
3672         if (!acc->name) {
3673                 if (relo->kind == BPF_FIELD_BYTE_OFFSET) {
3674                         *val = spec->bit_offset / 8;
3675                 } else if (relo->kind == BPF_FIELD_BYTE_SIZE) {
3676                         sz = btf__resolve_size(spec->btf, acc->type_id);
3677                         if (sz < 0)
3678                                 return -EINVAL;
3679                         *val = sz;
3680                 } else {
3681                         pr_warn("prog '%s': relo %d at insn #%d can't be applied to array access\n",
3682                                 bpf_program__title(prog, false),
3683                                 relo->kind, relo->insn_off / 8);
3684                         return -EINVAL;
3685                 }
3686                 if (validate)
3687                         *validate = true;
3688                 return 0;
3689         }
3690
3691         m = btf_members(t) + acc->idx;
3692         mt = skip_mods_and_typedefs(spec->btf, m->type, NULL);
3693         bit_off = spec->bit_offset;
3694         bit_sz = btf_member_bitfield_size(t, acc->idx);
3695
3696         bitfield = bit_sz > 0;
3697         if (bitfield) {
3698                 byte_sz = mt->size;
3699                 byte_off = bit_off / 8 / byte_sz * byte_sz;
3700                 /* figure out smallest int size necessary for bitfield load */
3701                 while (bit_off + bit_sz - byte_off * 8 > byte_sz * 8) {
3702                         if (byte_sz >= 8) {
3703                                 /* bitfield can't be read with 64-bit read */
3704                                 pr_warn("prog '%s': relo %d at insn #%d can't be satisfied for bitfield\n",
3705                                         bpf_program__title(prog, false),
3706                                         relo->kind, relo->insn_off / 8);
3707                                 return -E2BIG;
3708                         }
3709                         byte_sz *= 2;
3710                         byte_off = bit_off / 8 / byte_sz * byte_sz;
3711                 }
3712         } else {
3713                 sz = btf__resolve_size(spec->btf, m->type);
3714                 if (sz < 0)
3715                         return -EINVAL;
3716                 byte_sz = sz;
3717                 byte_off = spec->bit_offset / 8;
3718                 bit_sz = byte_sz * 8;
3719         }
3720
3721         /* for bitfields, all the relocatable aspects are ambiguous and we
3722          * might disagree with compiler, so turn off validation of expected
3723          * value, except for signedness
3724          */
3725         if (validate)
3726                 *validate = !bitfield;
3727
3728         switch (relo->kind) {
3729         case BPF_FIELD_BYTE_OFFSET:
3730                 *val = byte_off;
3731                 break;
3732         case BPF_FIELD_BYTE_SIZE:
3733                 *val = byte_sz;
3734                 break;
3735         case BPF_FIELD_SIGNED:
3736                 /* enums will be assumed unsigned */
3737                 *val = btf_is_enum(mt) ||
3738                        (btf_int_encoding(mt) & BTF_INT_SIGNED);
3739                 if (validate)
3740                         *validate = true; /* signedness is never ambiguous */
3741                 break;
3742         case BPF_FIELD_LSHIFT_U64:
3743 #if __BYTE_ORDER == __LITTLE_ENDIAN
3744                 *val = 64 - (bit_off + bit_sz - byte_off  * 8);
3745 #else
3746                 *val = (8 - byte_sz) * 8 + (bit_off - byte_off * 8);
3747 #endif
3748                 break;
3749         case BPF_FIELD_RSHIFT_U64:
3750                 *val = 64 - bit_sz;
3751                 if (validate)
3752                         *validate = true; /* right shift is never ambiguous */
3753                 break;
3754         case BPF_FIELD_EXISTS:
3755         default:
3756                 pr_warn("prog '%s': unknown relo %d at insn #%d\n",
3757                         bpf_program__title(prog, false),
3758                         relo->kind, relo->insn_off / 8);
3759                 return -EINVAL;
3760         }
3761
3762         return 0;
3763 }
3764
3765 /*
3766  * Patch relocatable BPF instruction.
3767  *
3768  * Patched value is determined by relocation kind and target specification.
3769  * For field existence relocation target spec will be NULL if field is not
3770  * found.
3771  * Expected insn->imm value is determined using relocation kind and local
3772  * spec, and is checked before patching instruction. If actual insn->imm value
3773  * is wrong, bail out with error.
3774  *
3775  * Currently three kinds of BPF instructions are supported:
3776  * 1. rX = <imm> (assignment with immediate operand);
3777  * 2. rX += <imm> (arithmetic operations with immediate operand);
3778  */
3779 static int bpf_core_reloc_insn(struct bpf_program *prog,
3780                                const struct bpf_field_reloc *relo,
3781                                const struct bpf_core_spec *local_spec,
3782                                const struct bpf_core_spec *targ_spec)
3783 {
3784         bool failed = false, validate = true;
3785         __u32 orig_val, new_val;
3786         struct bpf_insn *insn;
3787         int insn_idx, err;
3788         __u8 class;
3789
3790         if (relo->insn_off % sizeof(struct bpf_insn))
3791                 return -EINVAL;
3792         insn_idx = relo->insn_off / sizeof(struct bpf_insn);
3793
3794         if (relo->kind == BPF_FIELD_EXISTS) {
3795                 orig_val = 1; /* can't generate EXISTS relo w/o local field */
3796                 new_val = targ_spec ? 1 : 0;
3797         } else if (!targ_spec) {
3798                 failed = true;
3799                 new_val = (__u32)-1;
3800         } else {
3801                 err = bpf_core_calc_field_relo(prog, relo, local_spec,
3802                                                &orig_val, &validate);
3803                 if (err)
3804                         return err;
3805                 err = bpf_core_calc_field_relo(prog, relo, targ_spec,
3806                                                &new_val, NULL);
3807                 if (err)
3808                         return err;
3809         }
3810
3811         insn = &prog->insns[insn_idx];
3812         class = BPF_CLASS(insn->code);
3813
3814         switch (class) {
3815         case BPF_ALU:
3816         case BPF_ALU64:
3817                 if (BPF_SRC(insn->code) != BPF_K)
3818                         return -EINVAL;
3819                 if (!failed && validate && insn->imm != orig_val) {
3820                         pr_warn("prog '%s': unexpected insn #%d (ALU/ALU64) value: got %u, exp %u -> %u\n",
3821                                 bpf_program__title(prog, false), insn_idx,
3822                                 insn->imm, orig_val, new_val);
3823                         return -EINVAL;
3824                 }
3825                 orig_val = insn->imm;
3826                 insn->imm = new_val;
3827                 pr_debug("prog '%s': patched insn #%d (ALU/ALU64)%s imm %u -> %u\n",
3828                          bpf_program__title(prog, false), insn_idx,
3829                          failed ? " w/ failed reloc" : "", orig_val, new_val);
3830                 break;
3831         case BPF_LDX:
3832         case BPF_ST:
3833         case BPF_STX:
3834                 if (!failed && validate && insn->off != orig_val) {
3835                         pr_warn("prog '%s': unexpected insn #%d (LD/LDX/ST/STX) value: got %u, exp %u -> %u\n",
3836                                 bpf_program__title(prog, false), insn_idx,
3837                                 insn->off, orig_val, new_val);
3838                         return -EINVAL;
3839                 }
3840                 if (new_val > SHRT_MAX) {
3841                         pr_warn("prog '%s': insn #%d (LD/LDX/ST/STX) value too big: %u\n",
3842                                 bpf_program__title(prog, false), insn_idx,
3843                                 new_val);
3844                         return -ERANGE;
3845                 }
3846                 orig_val = insn->off;
3847                 insn->off = new_val;
3848                 pr_debug("prog '%s': patched insn #%d (LD/LDX/ST/STX)%s off %u -> %u\n",
3849                          bpf_program__title(prog, false), insn_idx,
3850                          failed ? " w/ failed reloc" : "", orig_val, new_val);
3851                 break;
3852         default:
3853                 pr_warn("prog '%s': trying to relocate unrecognized insn #%d, code:%x, src:%x, dst:%x, off:%x, imm:%x\n",
3854                         bpf_program__title(prog, false),
3855                         insn_idx, insn->code, insn->src_reg, insn->dst_reg,
3856                         insn->off, insn->imm);
3857                 return -EINVAL;
3858         }
3859
3860         return 0;
3861 }
3862
3863 static struct btf *btf_load_raw(const char *path)
3864 {
3865         struct btf *btf;
3866         size_t read_cnt;
3867         struct stat st;
3868         void *data;
3869         FILE *f;
3870
3871         if (stat(path, &st))
3872                 return ERR_PTR(-errno);
3873
3874         data = malloc(st.st_size);
3875         if (!data)
3876                 return ERR_PTR(-ENOMEM);
3877
3878         f = fopen(path, "rb");
3879         if (!f) {
3880                 btf = ERR_PTR(-errno);
3881                 goto cleanup;
3882         }
3883
3884         read_cnt = fread(data, 1, st.st_size, f);
3885         fclose(f);
3886         if (read_cnt < st.st_size) {
3887                 btf = ERR_PTR(-EBADF);
3888                 goto cleanup;
3889         }
3890
3891         btf = btf__new(data, read_cnt);
3892
3893 cleanup:
3894         free(data);
3895         return btf;
3896 }
3897
3898 /*
3899  * Probe few well-known locations for vmlinux kernel image and try to load BTF
3900  * data out of it to use for target BTF.
3901  */
3902 static struct btf *bpf_core_find_kernel_btf(void)
3903 {
3904         struct {
3905                 const char *path_fmt;
3906                 bool raw_btf;
3907         } locations[] = {
3908                 /* try canonical vmlinux BTF through sysfs first */
3909                 { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
3910                 /* fall back to trying to find vmlinux ELF on disk otherwise */
3911                 { "/boot/vmlinux-%1$s" },
3912                 { "/lib/modules/%1$s/vmlinux-%1$s" },
3913                 { "/lib/modules/%1$s/build/vmlinux" },
3914                 { "/usr/lib/modules/%1$s/kernel/vmlinux" },
3915                 { "/usr/lib/debug/boot/vmlinux-%1$s" },
3916                 { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
3917                 { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
3918         };
3919         char path[PATH_MAX + 1];
3920         struct utsname buf;
3921         struct btf *btf;
3922         int i;
3923
3924         uname(&buf);
3925
3926         for (i = 0; i < ARRAY_SIZE(locations); i++) {
3927                 snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
3928
3929                 if (access(path, R_OK))
3930                         continue;
3931
3932                 if (locations[i].raw_btf)
3933                         btf = btf_load_raw(path);
3934                 else
3935                         btf = btf__parse_elf(path, NULL);
3936
3937                 pr_debug("loading kernel BTF '%s': %ld\n",
3938                          path, IS_ERR(btf) ? PTR_ERR(btf) : 0);
3939                 if (IS_ERR(btf))
3940                         continue;
3941
3942                 return btf;
3943         }
3944
3945         pr_warn("failed to find valid kernel BTF\n");
3946         return ERR_PTR(-ESRCH);
3947 }
3948
3949 /* Output spec definition in the format:
3950  * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>,
3951  * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b
3952  */
3953 static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
3954 {
3955         const struct btf_type *t;
3956         const char *s;
3957         __u32 type_id;
3958         int i;
3959
3960         type_id = spec->spec[0].type_id;
3961         t = btf__type_by_id(spec->btf, type_id);
3962         s = btf__name_by_offset(spec->btf, t->name_off);
3963         libbpf_print(level, "[%u] %s + ", type_id, s);
3964
3965         for (i = 0; i < spec->raw_len; i++)
3966                 libbpf_print(level, "%d%s", spec->raw_spec[i],
3967                              i == spec->raw_len - 1 ? " => " : ":");
3968
3969         libbpf_print(level, "%u.%u @ &x",
3970                      spec->bit_offset / 8, spec->bit_offset % 8);
3971
3972         for (i = 0; i < spec->len; i++) {
3973                 if (spec->spec[i].name)
3974                         libbpf_print(level, ".%s", spec->spec[i].name);
3975                 else
3976                         libbpf_print(level, "[%u]", spec->spec[i].idx);
3977         }
3978
3979 }
3980
3981 static size_t bpf_core_hash_fn(const void *key, void *ctx)
3982 {
3983         return (size_t)key;
3984 }
3985
3986 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
3987 {
3988         return k1 == k2;
3989 }
3990
3991 static void *u32_as_hash_key(__u32 x)
3992 {
3993         return (void *)(uintptr_t)x;
3994 }
3995
3996 /*
3997  * CO-RE relocate single instruction.
3998  *
3999  * The outline and important points of the algorithm:
4000  * 1. For given local type, find corresponding candidate target types.
4001  *    Candidate type is a type with the same "essential" name, ignoring
4002  *    everything after last triple underscore (___). E.g., `sample`,
4003  *    `sample___flavor_one`, `sample___flavor_another_one`, are all candidates
4004  *    for each other. Names with triple underscore are referred to as
4005  *    "flavors" and are useful, among other things, to allow to
4006  *    specify/support incompatible variations of the same kernel struct, which
4007  *    might differ between different kernel versions and/or build
4008  *    configurations.
4009  *
4010  *    N.B. Struct "flavors" could be generated by bpftool's BTF-to-C
4011  *    converter, when deduplicated BTF of a kernel still contains more than
4012  *    one different types with the same name. In that case, ___2, ___3, etc
4013  *    are appended starting from second name conflict. But start flavors are
4014  *    also useful to be defined "locally", in BPF program, to extract same
4015  *    data from incompatible changes between different kernel
4016  *    versions/configurations. For instance, to handle field renames between
4017  *    kernel versions, one can use two flavors of the struct name with the
4018  *    same common name and use conditional relocations to extract that field,
4019  *    depending on target kernel version.
4020  * 2. For each candidate type, try to match local specification to this
4021  *    candidate target type. Matching involves finding corresponding
4022  *    high-level spec accessors, meaning that all named fields should match,
4023  *    as well as all array accesses should be within the actual bounds. Also,
4024  *    types should be compatible (see bpf_core_fields_are_compat for details).
4025  * 3. It is supported and expected that there might be multiple flavors
4026  *    matching the spec. As long as all the specs resolve to the same set of
4027  *    offsets across all candidates, there is no error. If there is any
4028  *    ambiguity, CO-RE relocation will fail. This is necessary to accomodate
4029  *    imprefection of BTF deduplication, which can cause slight duplication of
4030  *    the same BTF type, if some directly or indirectly referenced (by
4031  *    pointer) type gets resolved to different actual types in different
4032  *    object files. If such situation occurs, deduplicated BTF will end up
4033  *    with two (or more) structurally identical types, which differ only in
4034  *    types they refer to through pointer. This should be OK in most cases and
4035  *    is not an error.
4036  * 4. Candidate types search is performed by linearly scanning through all
4037  *    types in target BTF. It is anticipated that this is overall more
4038  *    efficient memory-wise and not significantly worse (if not better)
4039  *    CPU-wise compared to prebuilding a map from all local type names to
4040  *    a list of candidate type names. It's also sped up by caching resolved
4041  *    list of matching candidates per each local "root" type ID, that has at
4042  *    least one bpf_field_reloc associated with it. This list is shared
4043  *    between multiple relocations for the same type ID and is updated as some
4044  *    of the candidates are pruned due to structural incompatibility.
4045  */
4046 static int bpf_core_reloc_field(struct bpf_program *prog,
4047                                  const struct bpf_field_reloc *relo,
4048                                  int relo_idx,
4049                                  const struct btf *local_btf,
4050                                  const struct btf *targ_btf,
4051                                  struct hashmap *cand_cache)
4052 {
4053         const char *prog_name = bpf_program__title(prog, false);
4054         struct bpf_core_spec local_spec, cand_spec, targ_spec;
4055         const void *type_key = u32_as_hash_key(relo->type_id);
4056         const struct btf_type *local_type, *cand_type;
4057         const char *local_name, *cand_name;
4058         struct ids_vec *cand_ids;
4059         __u32 local_id, cand_id;
4060         const char *spec_str;
4061         int i, j, err;
4062
4063         local_id = relo->type_id;
4064         local_type = btf__type_by_id(local_btf, local_id);
4065         if (!local_type)
4066                 return -EINVAL;
4067
4068         local_name = btf__name_by_offset(local_btf, local_type->name_off);
4069         if (str_is_empty(local_name))
4070                 return -EINVAL;
4071
4072         spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
4073         if (str_is_empty(spec_str))
4074                 return -EINVAL;
4075
4076         err = bpf_core_spec_parse(local_btf, local_id, spec_str, &local_spec);
4077         if (err) {
4078                 pr_warn("prog '%s': relo #%d: parsing [%d] %s + %s failed: %d\n",
4079                         prog_name, relo_idx, local_id, local_name, spec_str,
4080                         err);
4081                 return -EINVAL;
4082         }
4083
4084         pr_debug("prog '%s': relo #%d: kind %d, spec is ", prog_name, relo_idx,
4085                  relo->kind);
4086         bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
4087         libbpf_print(LIBBPF_DEBUG, "\n");
4088
4089         if (!hashmap__find(cand_cache, type_key, (void **)&cand_ids)) {
4090                 cand_ids = bpf_core_find_cands(local_btf, local_id, targ_btf);
4091                 if (IS_ERR(cand_ids)) {
4092                         pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s: %ld",
4093                                 prog_name, relo_idx, local_id, local_name,
4094                                 PTR_ERR(cand_ids));
4095                         return PTR_ERR(cand_ids);
4096                 }
4097                 err = hashmap__set(cand_cache, type_key, cand_ids, NULL, NULL);
4098                 if (err) {
4099                         bpf_core_free_cands(cand_ids);
4100                         return err;
4101                 }
4102         }
4103
4104         for (i = 0, j = 0; i < cand_ids->len; i++) {
4105                 cand_id = cand_ids->data[i];
4106                 cand_type = btf__type_by_id(targ_btf, cand_id);
4107                 cand_name = btf__name_by_offset(targ_btf, cand_type->name_off);
4108
4109                 err = bpf_core_spec_match(&local_spec, targ_btf,
4110                                           cand_id, &cand_spec);
4111                 pr_debug("prog '%s': relo #%d: matching candidate #%d %s against spec ",
4112                          prog_name, relo_idx, i, cand_name);
4113                 bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec);
4114                 libbpf_print(LIBBPF_DEBUG, ": %d\n", err);
4115                 if (err < 0) {
4116                         pr_warn("prog '%s': relo #%d: matching error: %d\n",
4117                                 prog_name, relo_idx, err);
4118                         return err;
4119                 }
4120                 if (err == 0)
4121                         continue;
4122
4123                 if (j == 0) {
4124                         targ_spec = cand_spec;
4125                 } else if (cand_spec.bit_offset != targ_spec.bit_offset) {
4126                         /* if there are many candidates, they should all
4127                          * resolve to the same bit offset
4128                          */
4129                         pr_warn("prog '%s': relo #%d: offset ambiguity: %u != %u\n",
4130                                 prog_name, relo_idx, cand_spec.bit_offset,
4131                                 targ_spec.bit_offset);
4132                         return -EINVAL;
4133                 }
4134
4135                 cand_ids->data[j++] = cand_spec.spec[0].type_id;
4136         }
4137
4138         /*
4139          * For BPF_FIELD_EXISTS relo or when relaxed CO-RE reloc mode is
4140          * requested, it's expected that we might not find any candidates.
4141          * In this case, if field wasn't found in any candidate, the list of
4142          * candidates shouldn't change at all, we'll just handle relocating
4143          * appropriately, depending on relo's kind.
4144          */
4145         if (j > 0)
4146                 cand_ids->len = j;
4147
4148         if (j == 0 && !prog->obj->relaxed_core_relocs &&
4149             relo->kind != BPF_FIELD_EXISTS) {
4150                 pr_warn("prog '%s': relo #%d: no matching targets found for [%d] %s + %s\n",
4151                         prog_name, relo_idx, local_id, local_name, spec_str);
4152                 return -ESRCH;
4153         }
4154
4155         /* bpf_core_reloc_insn should know how to handle missing targ_spec */
4156         err = bpf_core_reloc_insn(prog, relo, &local_spec,
4157                                   j ? &targ_spec : NULL);
4158         if (err) {
4159                 pr_warn("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n",
4160                         prog_name, relo_idx, relo->insn_off, err);
4161                 return -EINVAL;
4162         }
4163
4164         return 0;
4165 }
4166
4167 static int
4168 bpf_core_reloc_fields(struct bpf_object *obj, const char *targ_btf_path)
4169 {
4170         const struct btf_ext_info_sec *sec;
4171         const struct bpf_field_reloc *rec;
4172         const struct btf_ext_info *seg;
4173         struct hashmap_entry *entry;
4174         struct hashmap *cand_cache = NULL;
4175         struct bpf_program *prog;
4176         struct btf *targ_btf;
4177         const char *sec_name;
4178         int i, err = 0;
4179
4180         if (targ_btf_path)
4181                 targ_btf = btf__parse_elf(targ_btf_path, NULL);
4182         else
4183                 targ_btf = bpf_core_find_kernel_btf();
4184         if (IS_ERR(targ_btf)) {
4185                 pr_warn("failed to get target BTF: %ld\n", PTR_ERR(targ_btf));
4186                 return PTR_ERR(targ_btf);
4187         }
4188
4189         cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
4190         if (IS_ERR(cand_cache)) {
4191                 err = PTR_ERR(cand_cache);
4192                 goto out;
4193         }
4194
4195         seg = &obj->btf_ext->field_reloc_info;
4196         for_each_btf_ext_sec(seg, sec) {
4197                 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
4198                 if (str_is_empty(sec_name)) {
4199                         err = -EINVAL;
4200                         goto out;
4201                 }
4202                 prog = bpf_object__find_program_by_title(obj, sec_name);
4203                 if (!prog) {
4204                         pr_warn("failed to find program '%s' for CO-RE offset relocation\n",
4205                                 sec_name);
4206                         err = -EINVAL;
4207                         goto out;
4208                 }
4209
4210                 pr_debug("prog '%s': performing %d CO-RE offset relocs\n",
4211                          sec_name, sec->num_info);
4212
4213                 for_each_btf_ext_rec(seg, sec, i, rec) {
4214                         err = bpf_core_reloc_field(prog, rec, i, obj->btf,
4215                                                    targ_btf, cand_cache);
4216                         if (err) {
4217                                 pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
4218                                         sec_name, i, err);
4219                                 goto out;
4220                         }
4221                 }
4222         }
4223
4224 out:
4225         btf__free(targ_btf);
4226         if (!IS_ERR_OR_NULL(cand_cache)) {
4227                 hashmap__for_each_entry(cand_cache, entry, i) {
4228                         bpf_core_free_cands(entry->value);
4229                 }
4230                 hashmap__free(cand_cache);
4231         }
4232         return err;
4233 }
4234
4235 static int
4236 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
4237 {
4238         int err = 0;
4239
4240         if (obj->btf_ext->field_reloc_info.len)
4241                 err = bpf_core_reloc_fields(obj, targ_btf_path);
4242
4243         return err;
4244 }
4245
4246 static int
4247 bpf_program__reloc_text(struct bpf_program *prog, struct bpf_object *obj,
4248                         struct reloc_desc *relo)
4249 {
4250         struct bpf_insn *insn, *new_insn;
4251         struct bpf_program *text;
4252         size_t new_cnt;
4253         int err;
4254
4255         if (prog->idx == obj->efile.text_shndx) {
4256                 pr_warn("relo in .text insn %d into off %d (insn #%d)\n",
4257                         relo->insn_idx, relo->sym_off, relo->sym_off / 8);
4258                 return -LIBBPF_ERRNO__RELOC;
4259         }
4260
4261         if (prog->main_prog_cnt == 0) {
4262                 text = bpf_object__find_prog_by_idx(obj, obj->efile.text_shndx);
4263                 if (!text) {
4264                         pr_warn("no .text section found yet relo into text exist\n");
4265                         return -LIBBPF_ERRNO__RELOC;
4266                 }
4267                 new_cnt = prog->insns_cnt + text->insns_cnt;
4268                 new_insn = reallocarray(prog->insns, new_cnt, sizeof(*insn));
4269                 if (!new_insn) {
4270                         pr_warn("oom in prog realloc\n");
4271                         return -ENOMEM;
4272                 }
4273                 prog->insns = new_insn;
4274
4275                 if (obj->btf_ext) {
4276                         err = bpf_program_reloc_btf_ext(prog, obj,
4277                                                         text->section_name,
4278                                                         prog->insns_cnt);
4279                         if (err)
4280                                 return err;
4281                 }
4282
4283                 memcpy(new_insn + prog->insns_cnt, text->insns,
4284                        text->insns_cnt * sizeof(*insn));
4285                 prog->main_prog_cnt = prog->insns_cnt;
4286                 prog->insns_cnt = new_cnt;
4287                 pr_debug("added %zd insn from %s to prog %s\n",
4288                          text->insns_cnt, text->section_name,
4289                          prog->section_name);
4290         }
4291         insn = &prog->insns[relo->insn_idx];
4292         insn->imm += relo->sym_off / 8 + prog->main_prog_cnt - relo->insn_idx;
4293         return 0;
4294 }
4295
4296 static int
4297 bpf_program__relocate(struct bpf_program *prog, struct bpf_object *obj)
4298 {
4299         int i, err;
4300
4301         if (!prog)
4302                 return 0;
4303
4304         if (obj->btf_ext) {
4305                 err = bpf_program_reloc_btf_ext(prog, obj,
4306                                                 prog->section_name, 0);
4307                 if (err)
4308                         return err;
4309         }
4310
4311         if (!prog->reloc_desc)
4312                 return 0;
4313
4314         for (i = 0; i < prog->nr_reloc; i++) {
4315                 struct reloc_desc *relo = &prog->reloc_desc[i];
4316                 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
4317
4318                 if (relo->insn_idx + 1 >= (int)prog->insns_cnt) {
4319                         pr_warn("relocation out of range: '%s'\n",
4320                                 prog->section_name);
4321                         return -LIBBPF_ERRNO__RELOC;
4322                 }
4323
4324                 switch (relo->type) {
4325                 case RELO_LD64:
4326                         insn[0].src_reg = BPF_PSEUDO_MAP_FD;
4327                         insn[0].imm = obj->maps[relo->map_idx].fd;
4328                         break;
4329                 case RELO_DATA:
4330                         insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
4331                         insn[1].imm = insn[0].imm + relo->sym_off;
4332                         insn[0].imm = obj->maps[relo->map_idx].fd;
4333                         break;
4334                 case RELO_EXTERN:
4335                         insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
4336                         insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
4337                         insn[1].imm = relo->sym_off;
4338                         break;
4339                 case RELO_CALL:
4340                         err = bpf_program__reloc_text(prog, obj, relo);
4341                         if (err)
4342                                 return err;
4343                         break;
4344                 default:
4345                         pr_warn("relo #%d: bad relo type %d\n", i, relo->type);
4346                         return -EINVAL;
4347                 }
4348         }
4349
4350         zfree(&prog->reloc_desc);
4351         prog->nr_reloc = 0;
4352         return 0;
4353 }
4354
4355 static int
4356 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
4357 {
4358         struct bpf_program *prog;
4359         size_t i;
4360         int err;
4361
4362         if (obj->btf_ext) {
4363                 err = bpf_object__relocate_core(obj, targ_btf_path);
4364                 if (err) {
4365                         pr_warn("failed to perform CO-RE relocations: %d\n",
4366                                 err);
4367                         return err;
4368                 }
4369         }
4370         for (i = 0; i < obj->nr_programs; i++) {
4371                 prog = &obj->programs[i];
4372
4373                 err = bpf_program__relocate(prog, obj);
4374                 if (err) {
4375                         pr_warn("failed to relocate '%s'\n", prog->section_name);
4376                         return err;
4377                 }
4378         }
4379         return 0;
4380 }
4381
4382 static int bpf_object__collect_reloc(struct bpf_object *obj)
4383 {
4384         int i, err;
4385
4386         if (!obj_elf_valid(obj)) {
4387                 pr_warn("Internal error: elf object is closed\n");
4388                 return -LIBBPF_ERRNO__INTERNAL;
4389         }
4390
4391         for (i = 0; i < obj->efile.nr_reloc_sects; i++) {
4392                 GElf_Shdr *shdr = &obj->efile.reloc_sects[i].shdr;
4393                 Elf_Data *data = obj->efile.reloc_sects[i].data;
4394                 int idx = shdr->sh_info;
4395                 struct bpf_program *prog;
4396
4397                 if (shdr->sh_type != SHT_REL) {
4398                         pr_warn("internal error at %d\n", __LINE__);
4399                         return -LIBBPF_ERRNO__INTERNAL;
4400                 }
4401
4402                 prog = bpf_object__find_prog_by_idx(obj, idx);
4403                 if (!prog) {
4404                         pr_warn("relocation failed: no section(%d)\n", idx);
4405                         return -LIBBPF_ERRNO__RELOC;
4406                 }
4407
4408                 err = bpf_program__collect_reloc(prog, shdr, data, obj);
4409                 if (err)
4410                         return err;
4411         }
4412         return 0;
4413 }
4414
4415 static int
4416 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
4417              char *license, __u32 kern_version, int *pfd)
4418 {
4419         struct bpf_load_program_attr load_attr;
4420         char *cp, errmsg[STRERR_BUFSIZE];
4421         int log_buf_size = BPF_LOG_BUF_SIZE;
4422         char *log_buf;
4423         int btf_fd, ret;
4424
4425         if (!insns || !insns_cnt)
4426                 return -EINVAL;
4427
4428         memset(&load_attr, 0, sizeof(struct bpf_load_program_attr));
4429         load_attr.prog_type = prog->type;
4430         load_attr.expected_attach_type = prog->expected_attach_type;
4431         if (prog->caps->name)
4432                 load_attr.name = prog->name;
4433         load_attr.insns = insns;
4434         load_attr.insns_cnt = insns_cnt;
4435         load_attr.license = license;
4436         if (prog->type == BPF_PROG_TYPE_TRACING) {
4437                 load_attr.attach_prog_fd = prog->attach_prog_fd;
4438                 load_attr.attach_btf_id = prog->attach_btf_id;
4439         } else {
4440                 load_attr.kern_version = kern_version;
4441                 load_attr.prog_ifindex = prog->prog_ifindex;
4442         }
4443         /* if .BTF.ext was loaded, kernel supports associated BTF for prog */
4444         if (prog->obj->btf_ext)
4445                 btf_fd = bpf_object__btf_fd(prog->obj);
4446         else
4447                 btf_fd = -1;
4448         load_attr.prog_btf_fd = btf_fd >= 0 ? btf_fd : 0;
4449         load_attr.func_info = prog->func_info;
4450         load_attr.func_info_rec_size = prog->func_info_rec_size;
4451         load_attr.func_info_cnt = prog->func_info_cnt;
4452         load_attr.line_info = prog->line_info;
4453         load_attr.line_info_rec_size = prog->line_info_rec_size;
4454         load_attr.line_info_cnt = prog->line_info_cnt;
4455         load_attr.log_level = prog->log_level;
4456         load_attr.prog_flags = prog->prog_flags;
4457
4458 retry_load:
4459         log_buf = malloc(log_buf_size);
4460         if (!log_buf)
4461                 pr_warn("Alloc log buffer for bpf loader error, continue without log\n");
4462
4463         ret = bpf_load_program_xattr(&load_attr, log_buf, log_buf_size);
4464
4465         if (ret >= 0) {
4466                 if (load_attr.log_level)
4467                         pr_debug("verifier log:\n%s", log_buf);
4468                 *pfd = ret;
4469                 ret = 0;
4470                 goto out;
4471         }
4472
4473         if (errno == ENOSPC) {
4474                 log_buf_size <<= 1;
4475                 free(log_buf);
4476                 goto retry_load;
4477         }
4478         ret = -errno;
4479         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
4480         pr_warn("load bpf program failed: %s\n", cp);
4481         pr_perm_msg(ret);
4482
4483         if (log_buf && log_buf[0] != '\0') {
4484                 ret = -LIBBPF_ERRNO__VERIFY;
4485                 pr_warn("-- BEGIN DUMP LOG ---\n");
4486                 pr_warn("\n%s\n", log_buf);
4487                 pr_warn("-- END LOG --\n");
4488         } else if (load_attr.insns_cnt >= BPF_MAXINSNS) {
4489                 pr_warn("Program too large (%zu insns), at most %d insns\n",
4490                         load_attr.insns_cnt, BPF_MAXINSNS);
4491                 ret = -LIBBPF_ERRNO__PROG2BIG;
4492         } else if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
4493                 /* Wrong program type? */
4494                 int fd;
4495
4496                 load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
4497                 load_attr.expected_attach_type = 0;
4498                 fd = bpf_load_program_xattr(&load_attr, NULL, 0);
4499                 if (fd >= 0) {
4500                         close(fd);
4501                         ret = -LIBBPF_ERRNO__PROGTYPE;
4502                         goto out;
4503                 }
4504         }
4505
4506 out:
4507         free(log_buf);
4508         return ret;
4509 }
4510
4511 static int libbpf_find_attach_btf_id(const char *name,
4512                                      enum bpf_attach_type attach_type,
4513                                      __u32 attach_prog_fd);
4514
4515 int bpf_program__load(struct bpf_program *prog, char *license, __u32 kern_ver)
4516 {
4517         int err = 0, fd, i, btf_id;
4518
4519         if (prog->type == BPF_PROG_TYPE_TRACING) {
4520                 btf_id = libbpf_find_attach_btf_id(prog->section_name,
4521                                                    prog->expected_attach_type,
4522                                                    prog->attach_prog_fd);
4523                 if (btf_id <= 0)
4524                         return btf_id;
4525                 prog->attach_btf_id = btf_id;
4526         }
4527
4528         if (prog->instances.nr < 0 || !prog->instances.fds) {
4529                 if (prog->preprocessor) {
4530                         pr_warn("Internal error: can't load program '%s'\n",
4531                                 prog->section_name);
4532                         return -LIBBPF_ERRNO__INTERNAL;
4533                 }
4534
4535                 prog->instances.fds = malloc(sizeof(int));
4536                 if (!prog->instances.fds) {
4537                         pr_warn("Not enough memory for BPF fds\n");
4538                         return -ENOMEM;
4539                 }
4540                 prog->instances.nr = 1;
4541                 prog->instances.fds[0] = -1;
4542         }
4543
4544         if (!prog->preprocessor) {
4545                 if (prog->instances.nr != 1) {
4546                         pr_warn("Program '%s' is inconsistent: nr(%d) != 1\n",
4547                                 prog->section_name, prog->instances.nr);
4548                 }
4549                 err = load_program(prog, prog->insns, prog->insns_cnt,
4550                                    license, kern_ver, &fd);
4551                 if (!err)
4552                         prog->instances.fds[0] = fd;
4553                 goto out;
4554         }
4555
4556         for (i = 0; i < prog->instances.nr; i++) {
4557                 struct bpf_prog_prep_result result;
4558                 bpf_program_prep_t preprocessor = prog->preprocessor;
4559
4560                 memset(&result, 0, sizeof(result));
4561                 err = preprocessor(prog, i, prog->insns,
4562                                    prog->insns_cnt, &result);
4563                 if (err) {
4564                         pr_warn("Preprocessing the %dth instance of program '%s' failed\n",
4565                                 i, prog->section_name);
4566                         goto out;
4567                 }
4568
4569                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
4570                         pr_debug("Skip loading the %dth instance of program '%s'\n",
4571                                  i, prog->section_name);
4572                         prog->instances.fds[i] = -1;
4573                         if (result.pfd)
4574                                 *result.pfd = -1;
4575                         continue;
4576                 }
4577
4578                 err = load_program(prog, result.new_insn_ptr,
4579                                    result.new_insn_cnt, license, kern_ver, &fd);
4580                 if (err) {
4581                         pr_warn("Loading the %dth instance of program '%s' failed\n",
4582                                 i, prog->section_name);
4583                         goto out;
4584                 }
4585
4586                 if (result.pfd)
4587                         *result.pfd = fd;
4588                 prog->instances.fds[i] = fd;
4589         }
4590 out:
4591         if (err)
4592                 pr_warn("failed to load program '%s'\n", prog->section_name);
4593         zfree(&prog->insns);
4594         prog->insns_cnt = 0;
4595         return err;
4596 }
4597
4598 static bool bpf_program__is_function_storage(const struct bpf_program *prog,
4599                                              const struct bpf_object *obj)
4600 {
4601         return prog->idx == obj->efile.text_shndx && obj->has_pseudo_calls;
4602 }
4603
4604 static int
4605 bpf_object__load_progs(struct bpf_object *obj, int log_level)
4606 {
4607         size_t i;
4608         int err;
4609
4610         for (i = 0; i < obj->nr_programs; i++) {
4611                 if (bpf_program__is_function_storage(&obj->programs[i], obj))
4612                         continue;
4613                 obj->programs[i].log_level |= log_level;
4614                 err = bpf_program__load(&obj->programs[i],
4615                                         obj->license,
4616                                         obj->kern_version);
4617                 if (err)
4618                         return err;
4619         }
4620         return 0;
4621 }
4622
4623 static struct bpf_object *
4624 __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
4625                    const struct bpf_object_open_opts *opts)
4626 {
4627         const char *obj_name, *kconfig;
4628         struct bpf_program *prog;
4629         struct bpf_object *obj;
4630         char tmp_name[64];
4631         int err;
4632
4633         if (elf_version(EV_CURRENT) == EV_NONE) {
4634                 pr_warn("failed to init libelf for %s\n",
4635                         path ? : "(mem buf)");
4636                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
4637         }
4638
4639         if (!OPTS_VALID(opts, bpf_object_open_opts))
4640                 return ERR_PTR(-EINVAL);
4641
4642         obj_name = OPTS_GET(opts, object_name, NULL);
4643         if (obj_buf) {
4644                 if (!obj_name) {
4645                         snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
4646                                  (unsigned long)obj_buf,
4647                                  (unsigned long)obj_buf_sz);
4648                         obj_name = tmp_name;
4649                 }
4650                 path = obj_name;
4651                 pr_debug("loading object '%s' from buffer\n", obj_name);
4652         }
4653
4654         obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
4655         if (IS_ERR(obj))
4656                 return obj;
4657
4658         obj->relaxed_core_relocs = OPTS_GET(opts, relaxed_core_relocs, false);
4659         kconfig = OPTS_GET(opts, kconfig, NULL);
4660         if (kconfig) {
4661                 obj->kconfig = strdup(kconfig);
4662                 if (!obj->kconfig)
4663                         return ERR_PTR(-ENOMEM);
4664         }
4665
4666         err = bpf_object__elf_init(obj);
4667         err = err ? : bpf_object__check_endianness(obj);
4668         err = err ? : bpf_object__elf_collect(obj);
4669         err = err ? : bpf_object__collect_externs(obj);
4670         err = err ? : bpf_object__finalize_btf(obj);
4671         err = err ? : bpf_object__init_maps(obj, opts);
4672         err = err ? : bpf_object__init_prog_names(obj);
4673         err = err ? : bpf_object__collect_reloc(obj);
4674         if (err)
4675                 goto out;
4676         bpf_object__elf_finish(obj);
4677
4678         bpf_object__for_each_program(prog, obj) {
4679                 enum bpf_prog_type prog_type;
4680                 enum bpf_attach_type attach_type;
4681
4682                 err = libbpf_prog_type_by_name(prog->section_name, &prog_type,
4683                                                &attach_type);
4684                 if (err == -ESRCH)
4685                         /* couldn't guess, but user might manually specify */
4686                         continue;
4687                 if (err)
4688                         goto out;
4689
4690                 bpf_program__set_type(prog, prog_type);
4691                 bpf_program__set_expected_attach_type(prog, attach_type);
4692                 if (prog_type == BPF_PROG_TYPE_TRACING)
4693                         prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
4694         }
4695
4696         return obj;
4697 out:
4698         bpf_object__close(obj);
4699         return ERR_PTR(err);
4700 }
4701
4702 static struct bpf_object *
4703 __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags)
4704 {
4705         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
4706                 .relaxed_maps = flags & MAPS_RELAX_COMPAT,
4707         );
4708
4709         /* param validation */
4710         if (!attr->file)
4711                 return NULL;
4712
4713         pr_debug("loading %s\n", attr->file);
4714         return __bpf_object__open(attr->file, NULL, 0, &opts);
4715 }
4716
4717 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
4718 {
4719         return __bpf_object__open_xattr(attr, 0);
4720 }
4721
4722 struct bpf_object *bpf_object__open(const char *path)
4723 {
4724         struct bpf_object_open_attr attr = {
4725                 .file           = path,
4726                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
4727         };
4728
4729         return bpf_object__open_xattr(&attr);
4730 }
4731
4732 struct bpf_object *
4733 bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
4734 {
4735         if (!path)
4736                 return ERR_PTR(-EINVAL);
4737
4738         pr_debug("loading %s\n", path);
4739
4740         return __bpf_object__open(path, NULL, 0, opts);
4741 }
4742
4743 struct bpf_object *
4744 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
4745                      const struct bpf_object_open_opts *opts)
4746 {
4747         if (!obj_buf || obj_buf_sz == 0)
4748                 return ERR_PTR(-EINVAL);
4749
4750         return __bpf_object__open(NULL, obj_buf, obj_buf_sz, opts);
4751 }
4752
4753 struct bpf_object *
4754 bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz,
4755                         const char *name)
4756 {
4757         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
4758                 .object_name = name,
4759                 /* wrong default, but backwards-compatible */
4760                 .relaxed_maps = true,
4761         );
4762
4763         /* returning NULL is wrong, but backwards-compatible */
4764         if (!obj_buf || obj_buf_sz == 0)
4765                 return NULL;
4766
4767         return bpf_object__open_mem(obj_buf, obj_buf_sz, &opts);
4768 }
4769
4770 int bpf_object__unload(struct bpf_object *obj)
4771 {
4772         size_t i;
4773
4774         if (!obj)
4775                 return -EINVAL;
4776
4777         for (i = 0; i < obj->nr_maps; i++)
4778                 zclose(obj->maps[i].fd);
4779
4780         for (i = 0; i < obj->nr_programs; i++)
4781                 bpf_program__unload(&obj->programs[i]);
4782
4783         return 0;
4784 }
4785
4786 static int bpf_object__sanitize_maps(struct bpf_object *obj)
4787 {
4788         struct bpf_map *m;
4789
4790         bpf_object__for_each_map(m, obj) {
4791                 if (!bpf_map__is_internal(m))
4792                         continue;
4793                 if (!obj->caps.global_data) {
4794                         pr_warn("kernel doesn't support global data\n");
4795                         return -ENOTSUP;
4796                 }
4797                 if (!obj->caps.array_mmap)
4798                         m->def.map_flags ^= BPF_F_MMAPABLE;
4799         }
4800
4801         return 0;
4802 }
4803
4804 static int bpf_object__resolve_externs(struct bpf_object *obj,
4805                                        const char *extra_kconfig)
4806 {
4807         bool need_config = false;
4808         struct extern_desc *ext;
4809         int err, i;
4810         void *data;
4811
4812         if (obj->nr_extern == 0)
4813                 return 0;
4814
4815         data = obj->maps[obj->kconfig_map_idx].mmaped;
4816
4817         for (i = 0; i < obj->nr_extern; i++) {
4818                 ext = &obj->externs[i];
4819
4820                 if (strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
4821                         void *ext_val = data + ext->data_off;
4822                         __u32 kver = get_kernel_version();
4823
4824                         if (!kver) {
4825                                 pr_warn("failed to get kernel version\n");
4826                                 return -EINVAL;
4827                         }
4828                         err = set_ext_value_num(ext, ext_val, kver);
4829                         if (err)
4830                                 return err;
4831                         pr_debug("extern %s=0x%x\n", ext->name, kver);
4832                 } else if (strncmp(ext->name, "CONFIG_", 7) == 0) {
4833                         need_config = true;
4834                 } else {
4835                         pr_warn("unrecognized extern '%s'\n", ext->name);
4836                         return -EINVAL;
4837                 }
4838         }
4839         if (need_config && extra_kconfig) {
4840                 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, data);
4841                 if (err)
4842                         return -EINVAL;
4843                 need_config = false;
4844                 for (i = 0; i < obj->nr_extern; i++) {
4845                         ext = &obj->externs[i];
4846                         if (!ext->is_set) {
4847                                 need_config = true;
4848                                 break;
4849                         }
4850                 }
4851         }
4852         if (need_config) {
4853                 err = bpf_object__read_kconfig_file(obj, data);
4854                 if (err)
4855                         return -EINVAL;
4856         }
4857         for (i = 0; i < obj->nr_extern; i++) {
4858                 ext = &obj->externs[i];
4859
4860                 if (!ext->is_set && !ext->is_weak) {
4861                         pr_warn("extern %s (strong) not resolved\n", ext->name);
4862                         return -ESRCH;
4863                 } else if (!ext->is_set) {
4864                         pr_debug("extern %s (weak) not resolved, defaulting to zero\n",
4865                                  ext->name);
4866                 }
4867         }
4868
4869         return 0;
4870 }
4871
4872 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
4873 {
4874         struct bpf_object *obj;
4875         int err, i;
4876
4877         if (!attr)
4878                 return -EINVAL;
4879         obj = attr->obj;
4880         if (!obj)
4881                 return -EINVAL;
4882
4883         if (obj->loaded) {
4884                 pr_warn("object should not be loaded twice\n");
4885                 return -EINVAL;
4886         }
4887
4888         obj->loaded = true;
4889
4890         err = bpf_object__probe_caps(obj);
4891         err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
4892         err = err ? : bpf_object__sanitize_and_load_btf(obj);
4893         err = err ? : bpf_object__sanitize_maps(obj);
4894         err = err ? : bpf_object__create_maps(obj);
4895         err = err ? : bpf_object__relocate(obj, attr->target_btf_path);
4896         err = err ? : bpf_object__load_progs(obj, attr->log_level);
4897         if (err)
4898                 goto out;
4899
4900         return 0;
4901 out:
4902         /* unpin any maps that were auto-pinned during load */
4903         for (i = 0; i < obj->nr_maps; i++)
4904                 if (obj->maps[i].pinned && !obj->maps[i].reused)
4905                         bpf_map__unpin(&obj->maps[i], NULL);
4906
4907         bpf_object__unload(obj);
4908         pr_warn("failed to load object '%s'\n", obj->path);
4909         return err;
4910 }
4911
4912 int bpf_object__load(struct bpf_object *obj)
4913 {
4914         struct bpf_object_load_attr attr = {
4915                 .obj = obj,
4916         };
4917
4918         return bpf_object__load_xattr(&attr);
4919 }
4920
4921 static int make_parent_dir(const char *path)
4922 {
4923         char *cp, errmsg[STRERR_BUFSIZE];
4924         char *dname, *dir;
4925         int err = 0;
4926
4927         dname = strdup(path);
4928         if (dname == NULL)
4929                 return -ENOMEM;
4930
4931         dir = dirname(dname);
4932         if (mkdir(dir, 0700) && errno != EEXIST)
4933                 err = -errno;
4934
4935         free(dname);
4936         if (err) {
4937                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
4938                 pr_warn("failed to mkdir %s: %s\n", path, cp);
4939         }
4940         return err;
4941 }
4942
4943 static int check_path(const char *path)
4944 {
4945         char *cp, errmsg[STRERR_BUFSIZE];
4946         struct statfs st_fs;
4947         char *dname, *dir;
4948         int err = 0;
4949
4950         if (path == NULL)
4951                 return -EINVAL;
4952
4953         dname = strdup(path);
4954         if (dname == NULL)
4955                 return -ENOMEM;
4956
4957         dir = dirname(dname);
4958         if (statfs(dir, &st_fs)) {
4959                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
4960                 pr_warn("failed to statfs %s: %s\n", dir, cp);
4961                 err = -errno;
4962         }
4963         free(dname);
4964
4965         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
4966                 pr_warn("specified path %s is not on BPF FS\n", path);
4967                 err = -EINVAL;
4968         }
4969
4970         return err;
4971 }
4972
4973 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
4974                               int instance)
4975 {
4976         char *cp, errmsg[STRERR_BUFSIZE];
4977         int err;
4978
4979         err = make_parent_dir(path);
4980         if (err)
4981                 return err;
4982
4983         err = check_path(path);
4984         if (err)
4985                 return err;
4986
4987         if (prog == NULL) {
4988                 pr_warn("invalid program pointer\n");
4989                 return -EINVAL;
4990         }
4991
4992         if (instance < 0 || instance >= prog->instances.nr) {
4993                 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
4994                         instance, prog->section_name, prog->instances.nr);
4995                 return -EINVAL;
4996         }
4997
4998         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
4999                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
5000                 pr_warn("failed to pin program: %s\n", cp);
5001                 return -errno;
5002         }
5003         pr_debug("pinned program '%s'\n", path);
5004
5005         return 0;
5006 }
5007
5008 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
5009                                 int instance)
5010 {
5011         int err;
5012
5013         err = check_path(path);
5014         if (err)
5015                 return err;
5016
5017         if (prog == NULL) {
5018                 pr_warn("invalid program pointer\n");
5019                 return -EINVAL;
5020         }
5021
5022         if (instance < 0 || instance >= prog->instances.nr) {
5023                 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
5024                         instance, prog->section_name, prog->instances.nr);
5025                 return -EINVAL;
5026         }
5027
5028         err = unlink(path);
5029         if (err != 0)
5030                 return -errno;
5031         pr_debug("unpinned program '%s'\n", path);
5032
5033         return 0;
5034 }
5035
5036 int bpf_program__pin(struct bpf_program *prog, const char *path)
5037 {
5038         int i, err;
5039
5040         err = make_parent_dir(path);
5041         if (err)
5042                 return err;
5043
5044         err = check_path(path);
5045         if (err)
5046                 return err;
5047
5048         if (prog == NULL) {
5049                 pr_warn("invalid program pointer\n");
5050                 return -EINVAL;
5051         }
5052
5053         if (prog->instances.nr <= 0) {
5054                 pr_warn("no instances of prog %s to pin\n",
5055                            prog->section_name);
5056                 return -EINVAL;
5057         }
5058
5059         if (prog->instances.nr == 1) {
5060                 /* don't create subdirs when pinning single instance */
5061                 return bpf_program__pin_instance(prog, path, 0);
5062         }
5063
5064         for (i = 0; i < prog->instances.nr; i++) {
5065                 char buf[PATH_MAX];
5066                 int len;
5067
5068                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
5069                 if (len < 0) {
5070                         err = -EINVAL;
5071                         goto err_unpin;
5072                 } else if (len >= PATH_MAX) {
5073                         err = -ENAMETOOLONG;
5074                         goto err_unpin;
5075                 }
5076
5077                 err = bpf_program__pin_instance(prog, buf, i);
5078                 if (err)
5079                         goto err_unpin;
5080         }
5081
5082         return 0;
5083
5084 err_unpin:
5085         for (i = i - 1; i >= 0; i--) {
5086                 char buf[PATH_MAX];
5087                 int len;
5088
5089                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
5090                 if (len < 0)
5091                         continue;
5092                 else if (len >= PATH_MAX)
5093                         continue;
5094
5095                 bpf_program__unpin_instance(prog, buf, i);
5096         }
5097
5098         rmdir(path);
5099
5100         return err;
5101 }
5102
5103 int bpf_program__unpin(struct bpf_program *prog, const char *path)
5104 {
5105         int i, err;
5106
5107         err = check_path(path);
5108         if (err)
5109                 return err;
5110
5111         if (prog == NULL) {
5112                 pr_warn("invalid program pointer\n");
5113                 return -EINVAL;
5114         }
5115
5116         if (prog->instances.nr <= 0) {
5117                 pr_warn("no instances of prog %s to pin\n",
5118                            prog->section_name);
5119                 return -EINVAL;
5120         }
5121
5122         if (prog->instances.nr == 1) {
5123                 /* don't create subdirs when pinning single instance */
5124                 return bpf_program__unpin_instance(prog, path, 0);
5125         }
5126
5127         for (i = 0; i < prog->instances.nr; i++) {
5128                 char buf[PATH_MAX];
5129                 int len;
5130
5131                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
5132                 if (len < 0)
5133                         return -EINVAL;
5134                 else if (len >= PATH_MAX)
5135                         return -ENAMETOOLONG;
5136
5137                 err = bpf_program__unpin_instance(prog, buf, i);
5138                 if (err)
5139                         return err;
5140         }
5141
5142         err = rmdir(path);
5143         if (err)
5144                 return -errno;
5145
5146         return 0;
5147 }
5148
5149 int bpf_map__pin(struct bpf_map *map, const char *path)
5150 {
5151         char *cp, errmsg[STRERR_BUFSIZE];
5152         int err;
5153
5154         if (map == NULL) {
5155                 pr_warn("invalid map pointer\n");
5156                 return -EINVAL;
5157         }
5158
5159         if (map->pin_path) {
5160                 if (path && strcmp(path, map->pin_path)) {
5161                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
5162                                 bpf_map__name(map), map->pin_path, path);
5163                         return -EINVAL;
5164                 } else if (map->pinned) {
5165                         pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
5166                                  bpf_map__name(map), map->pin_path);
5167                         return 0;
5168                 }
5169         } else {
5170                 if (!path) {
5171                         pr_warn("missing a path to pin map '%s' at\n",
5172                                 bpf_map__name(map));
5173                         return -EINVAL;
5174                 } else if (map->pinned) {
5175                         pr_warn("map '%s' already pinned\n", bpf_map__name(map));
5176                         return -EEXIST;
5177                 }
5178
5179                 map->pin_path = strdup(path);
5180                 if (!map->pin_path) {
5181                         err = -errno;
5182                         goto out_err;
5183                 }
5184         }
5185
5186         err = make_parent_dir(map->pin_path);
5187         if (err)
5188                 return err;
5189
5190         err = check_path(map->pin_path);
5191         if (err)
5192                 return err;
5193
5194         if (bpf_obj_pin(map->fd, map->pin_path)) {
5195                 err = -errno;
5196                 goto out_err;
5197         }
5198
5199         map->pinned = true;
5200         pr_debug("pinned map '%s'\n", map->pin_path);
5201
5202         return 0;
5203
5204 out_err:
5205         cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
5206         pr_warn("failed to pin map: %s\n", cp);
5207         return err;
5208 }
5209
5210 int bpf_map__unpin(struct bpf_map *map, const char *path)
5211 {
5212         int err;
5213
5214         if (map == NULL) {
5215                 pr_warn("invalid map pointer\n");
5216                 return -EINVAL;
5217         }
5218
5219         if (map->pin_path) {
5220                 if (path && strcmp(path, map->pin_path)) {
5221                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
5222                                 bpf_map__name(map), map->pin_path, path);
5223                         return -EINVAL;
5224                 }
5225                 path = map->pin_path;
5226         } else if (!path) {
5227                 pr_warn("no path to unpin map '%s' from\n",
5228                         bpf_map__name(map));
5229                 return -EINVAL;
5230         }
5231
5232         err = check_path(path);
5233         if (err)
5234                 return err;
5235
5236         err = unlink(path);
5237         if (err != 0)
5238                 return -errno;
5239
5240         map->pinned = false;
5241         pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
5242
5243         return 0;
5244 }
5245
5246 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
5247 {
5248         char *new = NULL;
5249
5250         if (path) {
5251                 new = strdup(path);
5252                 if (!new)
5253                         return -errno;
5254         }
5255
5256         free(map->pin_path);
5257         map->pin_path = new;
5258         return 0;
5259 }
5260
5261 const char *bpf_map__get_pin_path(const struct bpf_map *map)
5262 {
5263         return map->pin_path;
5264 }
5265
5266 bool bpf_map__is_pinned(const struct bpf_map *map)
5267 {
5268         return map->pinned;
5269 }
5270
5271 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
5272 {
5273         struct bpf_map *map;
5274         int err;
5275
5276         if (!obj)
5277                 return -ENOENT;
5278
5279         if (!obj->loaded) {
5280                 pr_warn("object not yet loaded; load it first\n");
5281                 return -ENOENT;
5282         }
5283
5284         bpf_object__for_each_map(map, obj) {
5285                 char *pin_path = NULL;
5286                 char buf[PATH_MAX];
5287
5288                 if (path) {
5289                         int len;
5290
5291                         len = snprintf(buf, PATH_MAX, "%s/%s", path,
5292                                        bpf_map__name(map));
5293                         if (len < 0) {
5294                                 err = -EINVAL;
5295                                 goto err_unpin_maps;
5296                         } else if (len >= PATH_MAX) {
5297                                 err = -ENAMETOOLONG;
5298                                 goto err_unpin_maps;
5299                         }
5300                         pin_path = buf;
5301                 } else if (!map->pin_path) {
5302                         continue;
5303                 }
5304
5305                 err = bpf_map__pin(map, pin_path);
5306                 if (err)
5307                         goto err_unpin_maps;
5308         }
5309
5310         return 0;
5311
5312 err_unpin_maps:
5313         while ((map = bpf_map__prev(map, obj))) {
5314                 if (!map->pin_path)
5315                         continue;
5316
5317                 bpf_map__unpin(map, NULL);
5318         }
5319
5320         return err;
5321 }
5322
5323 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
5324 {
5325         struct bpf_map *map;
5326         int err;
5327
5328         if (!obj)
5329                 return -ENOENT;
5330
5331         bpf_object__for_each_map(map, obj) {
5332                 char *pin_path = NULL;
5333                 char buf[PATH_MAX];
5334
5335                 if (path) {
5336                         int len;
5337
5338                         len = snprintf(buf, PATH_MAX, "%s/%s", path,
5339                                        bpf_map__name(map));
5340                         if (len < 0)
5341                                 return -EINVAL;
5342                         else if (len >= PATH_MAX)
5343                                 return -ENAMETOOLONG;
5344                         pin_path = buf;
5345                 } else if (!map->pin_path) {
5346                         continue;
5347                 }
5348
5349                 err = bpf_map__unpin(map, pin_path);
5350                 if (err)
5351                         return err;
5352         }
5353
5354         return 0;
5355 }
5356
5357 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
5358 {
5359         struct bpf_program *prog;
5360         int err;
5361
5362         if (!obj)
5363                 return -ENOENT;
5364
5365         if (!obj->loaded) {
5366                 pr_warn("object not yet loaded; load it first\n");
5367                 return -ENOENT;
5368         }
5369
5370         bpf_object__for_each_program(prog, obj) {
5371                 char buf[PATH_MAX];
5372                 int len;
5373
5374                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
5375                                prog->pin_name);
5376                 if (len < 0) {
5377                         err = -EINVAL;
5378                         goto err_unpin_programs;
5379                 } else if (len >= PATH_MAX) {
5380                         err = -ENAMETOOLONG;
5381                         goto err_unpin_programs;
5382                 }
5383
5384                 err = bpf_program__pin(prog, buf);
5385                 if (err)
5386                         goto err_unpin_programs;
5387         }
5388
5389         return 0;
5390
5391 err_unpin_programs:
5392         while ((prog = bpf_program__prev(prog, obj))) {
5393                 char buf[PATH_MAX];
5394                 int len;
5395
5396                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
5397                                prog->pin_name);
5398                 if (len < 0)
5399                         continue;
5400                 else if (len >= PATH_MAX)
5401                         continue;
5402
5403                 bpf_program__unpin(prog, buf);
5404         }
5405
5406         return err;
5407 }
5408
5409 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
5410 {
5411         struct bpf_program *prog;
5412         int err;
5413
5414         if (!obj)
5415                 return -ENOENT;
5416
5417         bpf_object__for_each_program(prog, obj) {
5418                 char buf[PATH_MAX];
5419                 int len;
5420
5421                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
5422                                prog->pin_name);
5423                 if (len < 0)
5424                         return -EINVAL;
5425                 else if (len >= PATH_MAX)
5426                         return -ENAMETOOLONG;
5427
5428                 err = bpf_program__unpin(prog, buf);
5429                 if (err)
5430                         return err;
5431         }
5432
5433         return 0;
5434 }
5435
5436 int bpf_object__pin(struct bpf_object *obj, const char *path)
5437 {
5438         int err;
5439
5440         err = bpf_object__pin_maps(obj, path);
5441         if (err)
5442                 return err;
5443
5444         err = bpf_object__pin_programs(obj, path);
5445         if (err) {
5446                 bpf_object__unpin_maps(obj, path);
5447                 return err;
5448         }
5449
5450         return 0;
5451 }
5452
5453 void bpf_object__close(struct bpf_object *obj)
5454 {
5455         size_t i;
5456
5457         if (!obj)
5458                 return;
5459
5460         if (obj->clear_priv)
5461                 obj->clear_priv(obj, obj->priv);
5462
5463         bpf_object__elf_finish(obj);
5464         bpf_object__unload(obj);
5465         btf__free(obj->btf);
5466         btf_ext__free(obj->btf_ext);
5467
5468         for (i = 0; i < obj->nr_maps; i++) {
5469                 struct bpf_map *map = &obj->maps[i];
5470
5471                 if (map->clear_priv)
5472                         map->clear_priv(map, map->priv);
5473                 map->priv = NULL;
5474                 map->clear_priv = NULL;
5475
5476                 if (map->mmaped) {
5477                         munmap(map->mmaped, bpf_map_mmap_sz(map));
5478                         map->mmaped = NULL;
5479                 }
5480
5481                 zfree(&map->name);
5482                 zfree(&map->pin_path);
5483         }
5484
5485         zfree(&obj->kconfig);
5486         zfree(&obj->externs);
5487         obj->nr_extern = 0;
5488
5489         zfree(&obj->maps);
5490         obj->nr_maps = 0;
5491
5492         if (obj->programs && obj->nr_programs) {
5493                 for (i = 0; i < obj->nr_programs; i++)
5494                         bpf_program__exit(&obj->programs[i]);
5495         }
5496         zfree(&obj->programs);
5497
5498         list_del(&obj->list);
5499         free(obj);
5500 }
5501
5502 struct bpf_object *
5503 bpf_object__next(struct bpf_object *prev)
5504 {
5505         struct bpf_object *next;
5506
5507         if (!prev)
5508                 next = list_first_entry(&bpf_objects_list,
5509                                         struct bpf_object,
5510                                         list);
5511         else
5512                 next = list_next_entry(prev, list);
5513
5514         /* Empty list is noticed here so don't need checking on entry. */
5515         if (&next->list == &bpf_objects_list)
5516                 return NULL;
5517
5518         return next;
5519 }
5520
5521 const char *bpf_object__name(const struct bpf_object *obj)
5522 {
5523         return obj ? obj->name : ERR_PTR(-EINVAL);
5524 }
5525
5526 unsigned int bpf_object__kversion(const struct bpf_object *obj)
5527 {
5528         return obj ? obj->kern_version : 0;
5529 }
5530
5531 struct btf *bpf_object__btf(const struct bpf_object *obj)
5532 {
5533         return obj ? obj->btf : NULL;
5534 }
5535
5536 int bpf_object__btf_fd(const struct bpf_object *obj)
5537 {
5538         return obj->btf ? btf__fd(obj->btf) : -1;
5539 }
5540
5541 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
5542                          bpf_object_clear_priv_t clear_priv)
5543 {
5544         if (obj->priv && obj->clear_priv)
5545                 obj->clear_priv(obj, obj->priv);
5546
5547         obj->priv = priv;
5548         obj->clear_priv = clear_priv;
5549         return 0;
5550 }
5551
5552 void *bpf_object__priv(const struct bpf_object *obj)
5553 {
5554         return obj ? obj->priv : ERR_PTR(-EINVAL);
5555 }
5556
5557 static struct bpf_program *
5558 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
5559                     bool forward)
5560 {
5561         size_t nr_programs = obj->nr_programs;
5562         ssize_t idx;
5563
5564         if (!nr_programs)
5565                 return NULL;
5566
5567         if (!p)
5568                 /* Iter from the beginning */
5569                 return forward ? &obj->programs[0] :
5570                         &obj->programs[nr_programs - 1];
5571
5572         if (p->obj != obj) {
5573                 pr_warn("error: program handler doesn't match object\n");
5574                 return NULL;
5575         }
5576
5577         idx = (p - obj->programs) + (forward ? 1 : -1);
5578         if (idx >= obj->nr_programs || idx < 0)
5579                 return NULL;
5580         return &obj->programs[idx];
5581 }
5582
5583 struct bpf_program *
5584 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
5585 {
5586         struct bpf_program *prog = prev;
5587
5588         do {
5589                 prog = __bpf_program__iter(prog, obj, true);
5590         } while (prog && bpf_program__is_function_storage(prog, obj));
5591
5592         return prog;
5593 }
5594
5595 struct bpf_program *
5596 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
5597 {
5598         struct bpf_program *prog = next;
5599
5600         do {
5601                 prog = __bpf_program__iter(prog, obj, false);
5602         } while (prog && bpf_program__is_function_storage(prog, obj));
5603
5604         return prog;
5605 }
5606
5607 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
5608                           bpf_program_clear_priv_t clear_priv)
5609 {
5610         if (prog->priv && prog->clear_priv)
5611                 prog->clear_priv(prog, prog->priv);
5612
5613         prog->priv = priv;
5614         prog->clear_priv = clear_priv;
5615         return 0;
5616 }
5617
5618 void *bpf_program__priv(const struct bpf_program *prog)
5619 {
5620         return prog ? prog->priv : ERR_PTR(-EINVAL);
5621 }
5622
5623 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
5624 {
5625         prog->prog_ifindex = ifindex;
5626 }
5627
5628 const char *bpf_program__name(const struct bpf_program *prog)
5629 {
5630         return prog->name;
5631 }
5632
5633 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
5634 {
5635         const char *title;
5636
5637         title = prog->section_name;
5638         if (needs_copy) {
5639                 title = strdup(title);
5640                 if (!title) {
5641                         pr_warn("failed to strdup program title\n");
5642                         return ERR_PTR(-ENOMEM);
5643                 }
5644         }
5645
5646         return title;
5647 }
5648
5649 int bpf_program__fd(const struct bpf_program *prog)
5650 {
5651         return bpf_program__nth_fd(prog, 0);
5652 }
5653
5654 size_t bpf_program__size(const struct bpf_program *prog)
5655 {
5656         return prog->insns_cnt * sizeof(struct bpf_insn);
5657 }
5658
5659 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
5660                           bpf_program_prep_t prep)
5661 {
5662         int *instances_fds;
5663
5664         if (nr_instances <= 0 || !prep)
5665                 return -EINVAL;
5666
5667         if (prog->instances.nr > 0 || prog->instances.fds) {
5668                 pr_warn("Can't set pre-processor after loading\n");
5669                 return -EINVAL;
5670         }
5671
5672         instances_fds = malloc(sizeof(int) * nr_instances);
5673         if (!instances_fds) {
5674                 pr_warn("alloc memory failed for fds\n");
5675                 return -ENOMEM;
5676         }
5677
5678         /* fill all fd with -1 */
5679         memset(instances_fds, -1, sizeof(int) * nr_instances);
5680
5681         prog->instances.nr = nr_instances;
5682         prog->instances.fds = instances_fds;
5683         prog->preprocessor = prep;
5684         return 0;
5685 }
5686
5687 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
5688 {
5689         int fd;
5690
5691         if (!prog)
5692                 return -EINVAL;
5693
5694         if (n >= prog->instances.nr || n < 0) {
5695                 pr_warn("Can't get the %dth fd from program %s: only %d instances\n",
5696                         n, prog->section_name, prog->instances.nr);
5697                 return -EINVAL;
5698         }
5699
5700         fd = prog->instances.fds[n];
5701         if (fd < 0) {
5702                 pr_warn("%dth instance of program '%s' is invalid\n",
5703                         n, prog->section_name);
5704                 return -ENOENT;
5705         }
5706
5707         return fd;
5708 }
5709
5710 enum bpf_prog_type bpf_program__get_type(struct bpf_program *prog)
5711 {
5712         return prog->type;
5713 }
5714
5715 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
5716 {
5717         prog->type = type;
5718 }
5719
5720 static bool bpf_program__is_type(const struct bpf_program *prog,
5721                                  enum bpf_prog_type type)
5722 {
5723         return prog ? (prog->type == type) : false;
5724 }
5725
5726 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                           \
5727 int bpf_program__set_##NAME(struct bpf_program *prog)           \
5728 {                                                               \
5729         if (!prog)                                              \
5730                 return -EINVAL;                                 \
5731         bpf_program__set_type(prog, TYPE);                      \
5732         return 0;                                               \
5733 }                                                               \
5734                                                                 \
5735 bool bpf_program__is_##NAME(const struct bpf_program *prog)     \
5736 {                                                               \
5737         return bpf_program__is_type(prog, TYPE);                \
5738 }                                                               \
5739
5740 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
5741 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
5742 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
5743 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
5744 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
5745 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
5746 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
5747 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
5748 BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING);
5749
5750 enum bpf_attach_type
5751 bpf_program__get_expected_attach_type(struct bpf_program *prog)
5752 {
5753         return prog->expected_attach_type;
5754 }
5755
5756 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
5757                                            enum bpf_attach_type type)
5758 {
5759         prog->expected_attach_type = type;
5760 }
5761
5762 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, is_attachable, btf, atype) \
5763         { string, sizeof(string) - 1, ptype, eatype, is_attachable, btf, atype }
5764
5765 /* Programs that can NOT be attached. */
5766 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0, 0)
5767
5768 /* Programs that can be attached. */
5769 #define BPF_APROG_SEC(string, ptype, atype) \
5770         BPF_PROG_SEC_IMPL(string, ptype, 0, 1, 0, atype)
5771
5772 /* Programs that must specify expected attach type at load time. */
5773 #define BPF_EAPROG_SEC(string, ptype, eatype) \
5774         BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, 0, eatype)
5775
5776 /* Programs that use BTF to identify attach point */
5777 #define BPF_PROG_BTF(string, ptype, eatype) \
5778         BPF_PROG_SEC_IMPL(string, ptype, eatype, 0, 1, 0)
5779
5780 /* Programs that can be attached but attach type can't be identified by section
5781  * name. Kept for backward compatibility.
5782  */
5783 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
5784
5785 #define SEC_DEF(sec_pfx, ptype, ...) {                                      \
5786         .sec = sec_pfx,                                                     \
5787         .len = sizeof(sec_pfx) - 1,                                         \
5788         .prog_type = BPF_PROG_TYPE_##ptype,                                 \
5789         __VA_ARGS__                                                         \
5790 }
5791
5792 struct bpf_sec_def;
5793
5794 typedef struct bpf_link *(*attach_fn_t)(const struct bpf_sec_def *sec,
5795                                         struct bpf_program *prog);
5796
5797 static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
5798                                       struct bpf_program *prog);
5799 static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
5800                                   struct bpf_program *prog);
5801 static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
5802                                       struct bpf_program *prog);
5803 static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
5804                                      struct bpf_program *prog);
5805
5806 struct bpf_sec_def {
5807         const char *sec;
5808         size_t len;
5809         enum bpf_prog_type prog_type;
5810         enum bpf_attach_type expected_attach_type;
5811         bool is_attachable;
5812         bool is_attach_btf;
5813         enum bpf_attach_type attach_type;
5814         attach_fn_t attach_fn;
5815 };
5816
5817 static const struct bpf_sec_def section_defs[] = {
5818         BPF_PROG_SEC("socket",                  BPF_PROG_TYPE_SOCKET_FILTER),
5819         BPF_PROG_SEC("sk_reuseport",            BPF_PROG_TYPE_SK_REUSEPORT),
5820         SEC_DEF("kprobe/", KPROBE,
5821                 .attach_fn = attach_kprobe),
5822         BPF_PROG_SEC("uprobe/",                 BPF_PROG_TYPE_KPROBE),
5823         SEC_DEF("kretprobe/", KPROBE,
5824                 .attach_fn = attach_kprobe),
5825         BPF_PROG_SEC("uretprobe/",              BPF_PROG_TYPE_KPROBE),
5826         BPF_PROG_SEC("classifier",              BPF_PROG_TYPE_SCHED_CLS),
5827         BPF_PROG_SEC("action",                  BPF_PROG_TYPE_SCHED_ACT),
5828         SEC_DEF("tracepoint/", TRACEPOINT,
5829                 .attach_fn = attach_tp),
5830         SEC_DEF("tp/", TRACEPOINT,
5831                 .attach_fn = attach_tp),
5832         SEC_DEF("raw_tracepoint/", RAW_TRACEPOINT,
5833                 .attach_fn = attach_raw_tp),
5834         SEC_DEF("raw_tp/", RAW_TRACEPOINT,
5835                 .attach_fn = attach_raw_tp),
5836         SEC_DEF("tp_btf/", TRACING,
5837                 .expected_attach_type = BPF_TRACE_RAW_TP,
5838                 .is_attach_btf = true,
5839                 .attach_fn = attach_trace),
5840         SEC_DEF("fentry/", TRACING,
5841                 .expected_attach_type = BPF_TRACE_FENTRY,
5842                 .is_attach_btf = true,
5843                 .attach_fn = attach_trace),
5844         SEC_DEF("fexit/", TRACING,
5845                 .expected_attach_type = BPF_TRACE_FEXIT,
5846                 .is_attach_btf = true,
5847                 .attach_fn = attach_trace),
5848         BPF_PROG_SEC("xdp",                     BPF_PROG_TYPE_XDP),
5849         BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
5850         BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
5851         BPF_PROG_SEC("lwt_out",                 BPF_PROG_TYPE_LWT_OUT),
5852         BPF_PROG_SEC("lwt_xmit",                BPF_PROG_TYPE_LWT_XMIT),
5853         BPF_PROG_SEC("lwt_seg6local",           BPF_PROG_TYPE_LWT_SEG6LOCAL),
5854         BPF_APROG_SEC("cgroup_skb/ingress",     BPF_PROG_TYPE_CGROUP_SKB,
5855                                                 BPF_CGROUP_INET_INGRESS),
5856         BPF_APROG_SEC("cgroup_skb/egress",      BPF_PROG_TYPE_CGROUP_SKB,
5857                                                 BPF_CGROUP_INET_EGRESS),
5858         BPF_APROG_COMPAT("cgroup/skb",          BPF_PROG_TYPE_CGROUP_SKB),
5859         BPF_APROG_SEC("cgroup/sock",            BPF_PROG_TYPE_CGROUP_SOCK,
5860                                                 BPF_CGROUP_INET_SOCK_CREATE),
5861         BPF_EAPROG_SEC("cgroup/post_bind4",     BPF_PROG_TYPE_CGROUP_SOCK,
5862                                                 BPF_CGROUP_INET4_POST_BIND),
5863         BPF_EAPROG_SEC("cgroup/post_bind6",     BPF_PROG_TYPE_CGROUP_SOCK,
5864                                                 BPF_CGROUP_INET6_POST_BIND),
5865         BPF_APROG_SEC("cgroup/dev",             BPF_PROG_TYPE_CGROUP_DEVICE,
5866                                                 BPF_CGROUP_DEVICE),
5867         BPF_APROG_SEC("sockops",                BPF_PROG_TYPE_SOCK_OPS,
5868                                                 BPF_CGROUP_SOCK_OPS),
5869         BPF_APROG_SEC("sk_skb/stream_parser",   BPF_PROG_TYPE_SK_SKB,
5870                                                 BPF_SK_SKB_STREAM_PARSER),
5871         BPF_APROG_SEC("sk_skb/stream_verdict",  BPF_PROG_TYPE_SK_SKB,
5872                                                 BPF_SK_SKB_STREAM_VERDICT),
5873         BPF_APROG_COMPAT("sk_skb",              BPF_PROG_TYPE_SK_SKB),
5874         BPF_APROG_SEC("sk_msg",                 BPF_PROG_TYPE_SK_MSG,
5875                                                 BPF_SK_MSG_VERDICT),
5876         BPF_APROG_SEC("lirc_mode2",             BPF_PROG_TYPE_LIRC_MODE2,
5877                                                 BPF_LIRC_MODE2),
5878         BPF_APROG_SEC("flow_dissector",         BPF_PROG_TYPE_FLOW_DISSECTOR,
5879                                                 BPF_FLOW_DISSECTOR),
5880         BPF_EAPROG_SEC("cgroup/bind4",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5881                                                 BPF_CGROUP_INET4_BIND),
5882         BPF_EAPROG_SEC("cgroup/bind6",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5883                                                 BPF_CGROUP_INET6_BIND),
5884         BPF_EAPROG_SEC("cgroup/connect4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5885                                                 BPF_CGROUP_INET4_CONNECT),
5886         BPF_EAPROG_SEC("cgroup/connect6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5887                                                 BPF_CGROUP_INET6_CONNECT),
5888         BPF_EAPROG_SEC("cgroup/sendmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5889                                                 BPF_CGROUP_UDP4_SENDMSG),
5890         BPF_EAPROG_SEC("cgroup/sendmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5891                                                 BPF_CGROUP_UDP6_SENDMSG),
5892         BPF_EAPROG_SEC("cgroup/recvmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5893                                                 BPF_CGROUP_UDP4_RECVMSG),
5894         BPF_EAPROG_SEC("cgroup/recvmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
5895                                                 BPF_CGROUP_UDP6_RECVMSG),
5896         BPF_EAPROG_SEC("cgroup/sysctl",         BPF_PROG_TYPE_CGROUP_SYSCTL,
5897                                                 BPF_CGROUP_SYSCTL),
5898         BPF_EAPROG_SEC("cgroup/getsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
5899                                                 BPF_CGROUP_GETSOCKOPT),
5900         BPF_EAPROG_SEC("cgroup/setsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
5901                                                 BPF_CGROUP_SETSOCKOPT),
5902 };
5903
5904 #undef BPF_PROG_SEC_IMPL
5905 #undef BPF_PROG_SEC
5906 #undef BPF_APROG_SEC
5907 #undef BPF_EAPROG_SEC
5908 #undef BPF_APROG_COMPAT
5909 #undef SEC_DEF
5910
5911 #define MAX_TYPE_NAME_SIZE 32
5912
5913 static const struct bpf_sec_def *find_sec_def(const char *sec_name)
5914 {
5915         int i, n = ARRAY_SIZE(section_defs);
5916
5917         for (i = 0; i < n; i++) {
5918                 if (strncmp(sec_name,
5919                             section_defs[i].sec, section_defs[i].len))
5920                         continue;
5921                 return &section_defs[i];
5922         }
5923         return NULL;
5924 }
5925
5926 static char *libbpf_get_type_names(bool attach_type)
5927 {
5928         int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
5929         char *buf;
5930
5931         buf = malloc(len);
5932         if (!buf)
5933                 return NULL;
5934
5935         buf[0] = '\0';
5936         /* Forge string buf with all available names */
5937         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
5938                 if (attach_type && !section_defs[i].is_attachable)
5939                         continue;
5940
5941                 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
5942                         free(buf);
5943                         return NULL;
5944                 }
5945                 strcat(buf, " ");
5946                 strcat(buf, section_defs[i].sec);
5947         }
5948
5949         return buf;
5950 }
5951
5952 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
5953                              enum bpf_attach_type *expected_attach_type)
5954 {
5955         const struct bpf_sec_def *sec_def;
5956         char *type_names;
5957
5958         if (!name)
5959                 return -EINVAL;
5960
5961         sec_def = find_sec_def(name);
5962         if (sec_def) {
5963                 *prog_type = sec_def->prog_type;
5964                 *expected_attach_type = sec_def->expected_attach_type;
5965                 return 0;
5966         }
5967
5968         pr_debug("failed to guess program type from ELF section '%s'\n", name);
5969         type_names = libbpf_get_type_names(false);
5970         if (type_names != NULL) {
5971                 pr_debug("supported section(type) names are:%s\n", type_names);
5972                 free(type_names);
5973         }
5974
5975         return -ESRCH;
5976 }
5977
5978 #define BTF_PREFIX "btf_trace_"
5979 int libbpf_find_vmlinux_btf_id(const char *name,
5980                                enum bpf_attach_type attach_type)
5981 {
5982         struct btf *btf = bpf_core_find_kernel_btf();
5983         char raw_tp_btf[128] = BTF_PREFIX;
5984         char *dst = raw_tp_btf + sizeof(BTF_PREFIX) - 1;
5985         const char *btf_name;
5986         int err = -EINVAL;
5987         __u32 kind;
5988
5989         if (IS_ERR(btf)) {
5990                 pr_warn("vmlinux BTF is not found\n");
5991                 return -EINVAL;
5992         }
5993
5994         if (attach_type == BPF_TRACE_RAW_TP) {
5995                 /* prepend "btf_trace_" prefix per kernel convention */
5996                 strncat(dst, name, sizeof(raw_tp_btf) - sizeof(BTF_PREFIX));
5997                 btf_name = raw_tp_btf;
5998                 kind = BTF_KIND_TYPEDEF;
5999         } else {
6000                 btf_name = name;
6001                 kind = BTF_KIND_FUNC;
6002         }
6003         err = btf__find_by_name_kind(btf, btf_name, kind);
6004         btf__free(btf);
6005         return err;
6006 }
6007
6008 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
6009 {
6010         struct bpf_prog_info_linear *info_linear;
6011         struct bpf_prog_info *info;
6012         struct btf *btf = NULL;
6013         int err = -EINVAL;
6014
6015         info_linear = bpf_program__get_prog_info_linear(attach_prog_fd, 0);
6016         if (IS_ERR_OR_NULL(info_linear)) {
6017                 pr_warn("failed get_prog_info_linear for FD %d\n",
6018                         attach_prog_fd);
6019                 return -EINVAL;
6020         }
6021         info = &info_linear->info;
6022         if (!info->btf_id) {
6023                 pr_warn("The target program doesn't have BTF\n");
6024                 goto out;
6025         }
6026         if (btf__get_from_id(info->btf_id, &btf)) {
6027                 pr_warn("Failed to get BTF of the program\n");
6028                 goto out;
6029         }
6030         err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
6031         btf__free(btf);
6032         if (err <= 0) {
6033                 pr_warn("%s is not found in prog's BTF\n", name);
6034                 goto out;
6035         }
6036 out:
6037         free(info_linear);
6038         return err;
6039 }
6040
6041 static int libbpf_find_attach_btf_id(const char *name,
6042                                      enum bpf_attach_type attach_type,
6043                                      __u32 attach_prog_fd)
6044 {
6045         int i, err;
6046
6047         if (!name)
6048                 return -EINVAL;
6049
6050         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
6051                 if (!section_defs[i].is_attach_btf)
6052                         continue;
6053                 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
6054                         continue;
6055                 if (attach_prog_fd)
6056                         err = libbpf_find_prog_btf_id(name + section_defs[i].len,
6057                                                       attach_prog_fd);
6058                 else
6059                         err = libbpf_find_vmlinux_btf_id(name + section_defs[i].len,
6060                                                          attach_type);
6061                 if (err <= 0)
6062                         pr_warn("%s is not found in vmlinux BTF\n", name);
6063                 return err;
6064         }
6065         pr_warn("failed to identify btf_id based on ELF section name '%s'\n", name);
6066         return -ESRCH;
6067 }
6068
6069 int libbpf_attach_type_by_name(const char *name,
6070                                enum bpf_attach_type *attach_type)
6071 {
6072         char *type_names;
6073         int i;
6074
6075         if (!name)
6076                 return -EINVAL;
6077
6078         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
6079                 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
6080                         continue;
6081                 if (!section_defs[i].is_attachable)
6082                         return -EINVAL;
6083                 *attach_type = section_defs[i].attach_type;
6084                 return 0;
6085         }
6086         pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
6087         type_names = libbpf_get_type_names(true);
6088         if (type_names != NULL) {
6089                 pr_debug("attachable section(type) names are:%s\n", type_names);
6090                 free(type_names);
6091         }
6092
6093         return -EINVAL;
6094 }
6095
6096 int bpf_map__fd(const struct bpf_map *map)
6097 {
6098         return map ? map->fd : -EINVAL;
6099 }
6100
6101 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
6102 {
6103         return map ? &map->def : ERR_PTR(-EINVAL);
6104 }
6105
6106 const char *bpf_map__name(const struct bpf_map *map)
6107 {
6108         return map ? map->name : NULL;
6109 }
6110
6111 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
6112 {
6113         return map ? map->btf_key_type_id : 0;
6114 }
6115
6116 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
6117 {
6118         return map ? map->btf_value_type_id : 0;
6119 }
6120
6121 int bpf_map__set_priv(struct bpf_map *map, void *priv,
6122                      bpf_map_clear_priv_t clear_priv)
6123 {
6124         if (!map)
6125                 return -EINVAL;
6126
6127         if (map->priv) {
6128                 if (map->clear_priv)
6129                         map->clear_priv(map, map->priv);
6130         }
6131
6132         map->priv = priv;
6133         map->clear_priv = clear_priv;
6134         return 0;
6135 }
6136
6137 void *bpf_map__priv(const struct bpf_map *map)
6138 {
6139         return map ? map->priv : ERR_PTR(-EINVAL);
6140 }
6141
6142 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
6143 {
6144         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
6145 }
6146
6147 bool bpf_map__is_internal(const struct bpf_map *map)
6148 {
6149         return map->libbpf_type != LIBBPF_MAP_UNSPEC;
6150 }
6151
6152 void bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
6153 {
6154         map->map_ifindex = ifindex;
6155 }
6156
6157 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
6158 {
6159         if (!bpf_map_type__is_map_in_map(map->def.type)) {
6160                 pr_warn("error: unsupported map type\n");
6161                 return -EINVAL;
6162         }
6163         if (map->inner_map_fd != -1) {
6164                 pr_warn("error: inner_map_fd already specified\n");
6165                 return -EINVAL;
6166         }
6167         map->inner_map_fd = fd;
6168         return 0;
6169 }
6170
6171 static struct bpf_map *
6172 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
6173 {
6174         ssize_t idx;
6175         struct bpf_map *s, *e;
6176
6177         if (!obj || !obj->maps)
6178                 return NULL;
6179
6180         s = obj->maps;
6181         e = obj->maps + obj->nr_maps;
6182
6183         if ((m < s) || (m >= e)) {
6184                 pr_warn("error in %s: map handler doesn't belong to object\n",
6185                          __func__);
6186                 return NULL;
6187         }
6188
6189         idx = (m - obj->maps) + i;
6190         if (idx >= obj->nr_maps || idx < 0)
6191                 return NULL;
6192         return &obj->maps[idx];
6193 }
6194
6195 struct bpf_map *
6196 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
6197 {
6198         if (prev == NULL)
6199                 return obj->maps;
6200
6201         return __bpf_map__iter(prev, obj, 1);
6202 }
6203
6204 struct bpf_map *
6205 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
6206 {
6207         if (next == NULL) {
6208                 if (!obj->nr_maps)
6209                         return NULL;
6210                 return obj->maps + obj->nr_maps - 1;
6211         }
6212
6213         return __bpf_map__iter(next, obj, -1);
6214 }
6215
6216 struct bpf_map *
6217 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
6218 {
6219         struct bpf_map *pos;
6220
6221         bpf_object__for_each_map(pos, obj) {
6222                 if (pos->name && !strcmp(pos->name, name))
6223                         return pos;
6224         }
6225         return NULL;
6226 }
6227
6228 int
6229 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
6230 {
6231         return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
6232 }
6233
6234 struct bpf_map *
6235 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
6236 {
6237         return ERR_PTR(-ENOTSUP);
6238 }
6239
6240 long libbpf_get_error(const void *ptr)
6241 {
6242         return PTR_ERR_OR_ZERO(ptr);
6243 }
6244
6245 int bpf_prog_load(const char *file, enum bpf_prog_type type,
6246                   struct bpf_object **pobj, int *prog_fd)
6247 {
6248         struct bpf_prog_load_attr attr;
6249
6250         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
6251         attr.file = file;
6252         attr.prog_type = type;
6253         attr.expected_attach_type = 0;
6254
6255         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
6256 }
6257
6258 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
6259                         struct bpf_object **pobj, int *prog_fd)
6260 {
6261         struct bpf_object_open_attr open_attr = {};
6262         struct bpf_program *prog, *first_prog = NULL;
6263         struct bpf_object *obj;
6264         struct bpf_map *map;
6265         int err;
6266
6267         if (!attr)
6268                 return -EINVAL;
6269         if (!attr->file)
6270                 return -EINVAL;
6271
6272         open_attr.file = attr->file;
6273         open_attr.prog_type = attr->prog_type;
6274
6275         obj = bpf_object__open_xattr(&open_attr);
6276         if (IS_ERR_OR_NULL(obj))
6277                 return -ENOENT;
6278
6279         bpf_object__for_each_program(prog, obj) {
6280                 enum bpf_attach_type attach_type = attr->expected_attach_type;
6281                 /*
6282                  * to preserve backwards compatibility, bpf_prog_load treats
6283                  * attr->prog_type, if specified, as an override to whatever
6284                  * bpf_object__open guessed
6285                  */
6286                 if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) {
6287                         bpf_program__set_type(prog, attr->prog_type);
6288                         bpf_program__set_expected_attach_type(prog,
6289                                                               attach_type);
6290                 }
6291                 if (bpf_program__get_type(prog) == BPF_PROG_TYPE_UNSPEC) {
6292                         /*
6293                          * we haven't guessed from section name and user
6294                          * didn't provide a fallback type, too bad...
6295                          */
6296                         bpf_object__close(obj);
6297                         return -EINVAL;
6298                 }
6299
6300                 prog->prog_ifindex = attr->ifindex;
6301                 prog->log_level = attr->log_level;
6302                 prog->prog_flags = attr->prog_flags;
6303                 if (!first_prog)
6304                         first_prog = prog;
6305         }
6306
6307         bpf_object__for_each_map(map, obj) {
6308                 if (!bpf_map__is_offload_neutral(map))
6309                         map->map_ifindex = attr->ifindex;
6310         }
6311
6312         if (!first_prog) {
6313                 pr_warn("object file doesn't contain bpf program\n");
6314                 bpf_object__close(obj);
6315                 return -ENOENT;
6316         }
6317
6318         err = bpf_object__load(obj);
6319         if (err) {
6320                 bpf_object__close(obj);
6321                 return -EINVAL;
6322         }
6323
6324         *pobj = obj;
6325         *prog_fd = bpf_program__fd(first_prog);
6326         return 0;
6327 }
6328
6329 struct bpf_link {
6330         int (*detach)(struct bpf_link *link);
6331         int (*destroy)(struct bpf_link *link);
6332         bool disconnected;
6333 };
6334
6335 /* Release "ownership" of underlying BPF resource (typically, BPF program
6336  * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
6337  * link, when destructed through bpf_link__destroy() call won't attempt to
6338  * detach/unregisted that BPF resource. This is useful in situations where,
6339  * say, attached BPF program has to outlive userspace program that attached it
6340  * in the system. Depending on type of BPF program, though, there might be
6341  * additional steps (like pinning BPF program in BPF FS) necessary to ensure
6342  * exit of userspace program doesn't trigger automatic detachment and clean up
6343  * inside the kernel.
6344  */
6345 void bpf_link__disconnect(struct bpf_link *link)
6346 {
6347         link->disconnected = true;
6348 }
6349
6350 int bpf_link__destroy(struct bpf_link *link)
6351 {
6352         int err = 0;
6353
6354         if (!link)
6355                 return 0;
6356
6357         if (!link->disconnected && link->detach)
6358                 err = link->detach(link);
6359         if (link->destroy)
6360                 link->destroy(link);
6361         free(link);
6362
6363         return err;
6364 }
6365
6366 struct bpf_link_fd {
6367         struct bpf_link link; /* has to be at the top of struct */
6368         int fd; /* hook FD */
6369 };
6370
6371 static int bpf_link__detach_perf_event(struct bpf_link *link)
6372 {
6373         struct bpf_link_fd *l = (void *)link;
6374         int err;
6375
6376         err = ioctl(l->fd, PERF_EVENT_IOC_DISABLE, 0);
6377         if (err)
6378                 err = -errno;
6379
6380         close(l->fd);
6381         return err;
6382 }
6383
6384 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
6385                                                 int pfd)
6386 {
6387         char errmsg[STRERR_BUFSIZE];
6388         struct bpf_link_fd *link;
6389         int prog_fd, err;
6390
6391         if (pfd < 0) {
6392                 pr_warn("program '%s': invalid perf event FD %d\n",
6393                         bpf_program__title(prog, false), pfd);
6394                 return ERR_PTR(-EINVAL);
6395         }
6396         prog_fd = bpf_program__fd(prog);
6397         if (prog_fd < 0) {
6398                 pr_warn("program '%s': can't attach BPF program w/o FD (did you load it?)\n",
6399                         bpf_program__title(prog, false));
6400                 return ERR_PTR(-EINVAL);
6401         }
6402
6403         link = calloc(1, sizeof(*link));
6404         if (!link)
6405                 return ERR_PTR(-ENOMEM);
6406         link->link.detach = &bpf_link__detach_perf_event;
6407         link->fd = pfd;
6408
6409         if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
6410                 err = -errno;
6411                 free(link);
6412                 pr_warn("program '%s': failed to attach to pfd %d: %s\n",
6413                         bpf_program__title(prog, false), pfd,
6414                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
6415                 return ERR_PTR(err);
6416         }
6417         if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
6418                 err = -errno;
6419                 free(link);
6420                 pr_warn("program '%s': failed to enable pfd %d: %s\n",
6421                         bpf_program__title(prog, false), pfd,
6422                            libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
6423                 return ERR_PTR(err);
6424         }
6425         return (struct bpf_link *)link;
6426 }
6427
6428 /*
6429  * this function is expected to parse integer in the range of [0, 2^31-1] from
6430  * given file using scanf format string fmt. If actual parsed value is
6431  * negative, the result might be indistinguishable from error
6432  */
6433 static int parse_uint_from_file(const char *file, const char *fmt)
6434 {
6435         char buf[STRERR_BUFSIZE];
6436         int err, ret;
6437         FILE *f;
6438
6439         f = fopen(file, "r");
6440         if (!f) {
6441                 err = -errno;
6442                 pr_debug("failed to open '%s': %s\n", file,
6443                          libbpf_strerror_r(err, buf, sizeof(buf)));
6444                 return err;
6445         }
6446         err = fscanf(f, fmt, &ret);
6447         if (err != 1) {
6448                 err = err == EOF ? -EIO : -errno;
6449                 pr_debug("failed to parse '%s': %s\n", file,
6450                         libbpf_strerror_r(err, buf, sizeof(buf)));
6451                 fclose(f);
6452                 return err;
6453         }
6454         fclose(f);
6455         return ret;
6456 }
6457
6458 static int determine_kprobe_perf_type(void)
6459 {
6460         const char *file = "/sys/bus/event_source/devices/kprobe/type";
6461
6462         return parse_uint_from_file(file, "%d\n");
6463 }
6464
6465 static int determine_uprobe_perf_type(void)
6466 {
6467         const char *file = "/sys/bus/event_source/devices/uprobe/type";
6468
6469         return parse_uint_from_file(file, "%d\n");
6470 }
6471
6472 static int determine_kprobe_retprobe_bit(void)
6473 {
6474         const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
6475
6476         return parse_uint_from_file(file, "config:%d\n");
6477 }
6478
6479 static int determine_uprobe_retprobe_bit(void)
6480 {
6481         const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
6482
6483         return parse_uint_from_file(file, "config:%d\n");
6484 }
6485
6486 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
6487                                  uint64_t offset, int pid)
6488 {
6489         struct perf_event_attr attr = {};
6490         char errmsg[STRERR_BUFSIZE];
6491         int type, pfd, err;
6492
6493         type = uprobe ? determine_uprobe_perf_type()
6494                       : determine_kprobe_perf_type();
6495         if (type < 0) {
6496                 pr_warn("failed to determine %s perf type: %s\n",
6497                         uprobe ? "uprobe" : "kprobe",
6498                         libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
6499                 return type;
6500         }
6501         if (retprobe) {
6502                 int bit = uprobe ? determine_uprobe_retprobe_bit()
6503                                  : determine_kprobe_retprobe_bit();
6504
6505                 if (bit < 0) {
6506                         pr_warn("failed to determine %s retprobe bit: %s\n",
6507                                 uprobe ? "uprobe" : "kprobe",
6508                                 libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
6509                         return bit;
6510                 }
6511                 attr.config |= 1 << bit;
6512         }
6513         attr.size = sizeof(attr);
6514         attr.type = type;
6515         attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
6516         attr.config2 = offset;           /* kprobe_addr or probe_offset */
6517
6518         /* pid filter is meaningful only for uprobes */
6519         pfd = syscall(__NR_perf_event_open, &attr,
6520                       pid < 0 ? -1 : pid /* pid */,
6521                       pid == -1 ? 0 : -1 /* cpu */,
6522                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
6523         if (pfd < 0) {
6524                 err = -errno;
6525                 pr_warn("%s perf_event_open() failed: %s\n",
6526                         uprobe ? "uprobe" : "kprobe",
6527                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
6528                 return err;
6529         }
6530         return pfd;
6531 }
6532
6533 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
6534                                             bool retprobe,
6535                                             const char *func_name)
6536 {
6537         char errmsg[STRERR_BUFSIZE];
6538         struct bpf_link *link;
6539         int pfd, err;
6540
6541         pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
6542                                     0 /* offset */, -1 /* pid */);
6543         if (pfd < 0) {
6544                 pr_warn("program '%s': failed to create %s '%s' perf event: %s\n",
6545                         bpf_program__title(prog, false),
6546                         retprobe ? "kretprobe" : "kprobe", func_name,
6547                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
6548                 return ERR_PTR(pfd);
6549         }
6550         link = bpf_program__attach_perf_event(prog, pfd);
6551         if (IS_ERR(link)) {
6552                 close(pfd);
6553                 err = PTR_ERR(link);
6554                 pr_warn("program '%s': failed to attach to %s '%s': %s\n",
6555                         bpf_program__title(prog, false),
6556                         retprobe ? "kretprobe" : "kprobe", func_name,
6557                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
6558                 return link;
6559         }
6560         return link;
6561 }
6562
6563 static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
6564                                       struct bpf_program *prog)
6565 {
6566         const char *func_name;
6567         bool retprobe;
6568
6569         func_name = bpf_program__title(prog, false) + sec->len;
6570         retprobe = strcmp(sec->sec, "kretprobe/") == 0;
6571
6572         return bpf_program__attach_kprobe(prog, retprobe, func_name);
6573 }
6574
6575 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
6576                                             bool retprobe, pid_t pid,
6577                                             const char *binary_path,
6578                                             size_t func_offset)
6579 {
6580         char errmsg[STRERR_BUFSIZE];
6581         struct bpf_link *link;
6582         int pfd, err;
6583
6584         pfd = perf_event_open_probe(true /* uprobe */, retprobe,
6585                                     binary_path, func_offset, pid);
6586         if (pfd < 0) {
6587                 pr_warn("program '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
6588                         bpf_program__title(prog, false),
6589                         retprobe ? "uretprobe" : "uprobe",
6590                         binary_path, func_offset,
6591                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
6592                 return ERR_PTR(pfd);
6593         }
6594         link = bpf_program__attach_perf_event(prog, pfd);
6595         if (IS_ERR(link)) {
6596                 close(pfd);
6597                 err = PTR_ERR(link);
6598                 pr_warn("program '%s': failed to attach to %s '%s:0x%zx': %s\n",
6599                         bpf_program__title(prog, false),
6600                         retprobe ? "uretprobe" : "uprobe",
6601                         binary_path, func_offset,
6602                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
6603                 return link;
6604         }
6605         return link;
6606 }
6607
6608 static int determine_tracepoint_id(const char *tp_category,
6609                                    const char *tp_name)
6610 {
6611         char file[PATH_MAX];
6612         int ret;
6613
6614         ret = snprintf(file, sizeof(file),
6615                        "/sys/kernel/debug/tracing/events/%s/%s/id",
6616                        tp_category, tp_name);
6617         if (ret < 0)
6618                 return -errno;
6619         if (ret >= sizeof(file)) {
6620                 pr_debug("tracepoint %s/%s path is too long\n",
6621                          tp_category, tp_name);
6622                 return -E2BIG;
6623         }
6624         return parse_uint_from_file(file, "%d\n");
6625 }
6626
6627 static int perf_event_open_tracepoint(const char *tp_category,
6628                                       const char *tp_name)
6629 {
6630         struct perf_event_attr attr = {};
6631         char errmsg[STRERR_BUFSIZE];
6632         int tp_id, pfd, err;
6633
6634         tp_id = determine_tracepoint_id(tp_category, tp_name);
6635         if (tp_id < 0) {
6636                 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
6637                         tp_category, tp_name,
6638                         libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
6639                 return tp_id;
6640         }
6641
6642         attr.type = PERF_TYPE_TRACEPOINT;
6643         attr.size = sizeof(attr);
6644         attr.config = tp_id;
6645
6646         pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
6647                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
6648         if (pfd < 0) {
6649                 err = -errno;
6650                 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
6651                         tp_category, tp_name,
6652                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
6653                 return err;
6654         }
6655         return pfd;
6656 }
6657
6658 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
6659                                                 const char *tp_category,
6660                                                 const char *tp_name)
6661 {
6662         char errmsg[STRERR_BUFSIZE];
6663         struct bpf_link *link;
6664         int pfd, err;
6665
6666         pfd = perf_event_open_tracepoint(tp_category, tp_name);
6667         if (pfd < 0) {
6668                 pr_warn("program '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
6669                         bpf_program__title(prog, false),
6670                         tp_category, tp_name,
6671                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
6672                 return ERR_PTR(pfd);
6673         }
6674         link = bpf_program__attach_perf_event(prog, pfd);
6675         if (IS_ERR(link)) {
6676                 close(pfd);
6677                 err = PTR_ERR(link);
6678                 pr_warn("program '%s': failed to attach to tracepoint '%s/%s': %s\n",
6679                         bpf_program__title(prog, false),
6680                         tp_category, tp_name,
6681                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
6682                 return link;
6683         }
6684         return link;
6685 }
6686
6687 static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
6688                                   struct bpf_program *prog)
6689 {
6690         char *sec_name, *tp_cat, *tp_name;
6691         struct bpf_link *link;
6692
6693         sec_name = strdup(bpf_program__title(prog, false));
6694         if (!sec_name)
6695                 return ERR_PTR(-ENOMEM);
6696
6697         /* extract "tp/<category>/<name>" */
6698         tp_cat = sec_name + sec->len;
6699         tp_name = strchr(tp_cat, '/');
6700         if (!tp_name) {
6701                 link = ERR_PTR(-EINVAL);
6702                 goto out;
6703         }
6704         *tp_name = '\0';
6705         tp_name++;
6706
6707         link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
6708 out:
6709         free(sec_name);
6710         return link;
6711 }
6712
6713 static int bpf_link__detach_fd(struct bpf_link *link)
6714 {
6715         struct bpf_link_fd *l = (void *)link;
6716
6717         return close(l->fd);
6718 }
6719
6720 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
6721                                                     const char *tp_name)
6722 {
6723         char errmsg[STRERR_BUFSIZE];
6724         struct bpf_link_fd *link;
6725         int prog_fd, pfd;
6726
6727         prog_fd = bpf_program__fd(prog);
6728         if (prog_fd < 0) {
6729                 pr_warn("program '%s': can't attach before loaded\n",
6730                         bpf_program__title(prog, false));
6731                 return ERR_PTR(-EINVAL);
6732         }
6733
6734         link = calloc(1, sizeof(*link));
6735         if (!link)
6736                 return ERR_PTR(-ENOMEM);
6737         link->link.detach = &bpf_link__detach_fd;
6738
6739         pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
6740         if (pfd < 0) {
6741                 pfd = -errno;
6742                 free(link);
6743                 pr_warn("program '%s': failed to attach to raw tracepoint '%s': %s\n",
6744                         bpf_program__title(prog, false), tp_name,
6745                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
6746                 return ERR_PTR(pfd);
6747         }
6748         link->fd = pfd;
6749         return (struct bpf_link *)link;
6750 }
6751
6752 static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
6753                                       struct bpf_program *prog)
6754 {
6755         const char *tp_name = bpf_program__title(prog, false) + sec->len;
6756
6757         return bpf_program__attach_raw_tracepoint(prog, tp_name);
6758 }
6759
6760 struct bpf_link *bpf_program__attach_trace(struct bpf_program *prog)
6761 {
6762         char errmsg[STRERR_BUFSIZE];
6763         struct bpf_link_fd *link;
6764         int prog_fd, pfd;
6765
6766         prog_fd = bpf_program__fd(prog);
6767         if (prog_fd < 0) {
6768                 pr_warn("program '%s': can't attach before loaded\n",
6769                         bpf_program__title(prog, false));
6770                 return ERR_PTR(-EINVAL);
6771         }
6772
6773         link = calloc(1, sizeof(*link));
6774         if (!link)
6775                 return ERR_PTR(-ENOMEM);
6776         link->link.detach = &bpf_link__detach_fd;
6777
6778         pfd = bpf_raw_tracepoint_open(NULL, prog_fd);
6779         if (pfd < 0) {
6780                 pfd = -errno;
6781                 free(link);
6782                 pr_warn("program '%s': failed to attach to trace: %s\n",
6783                         bpf_program__title(prog, false),
6784                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
6785                 return ERR_PTR(pfd);
6786         }
6787         link->fd = pfd;
6788         return (struct bpf_link *)link;
6789 }
6790
6791 static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
6792                                      struct bpf_program *prog)
6793 {
6794         return bpf_program__attach_trace(prog);
6795 }
6796
6797 struct bpf_link *bpf_program__attach(struct bpf_program *prog)
6798 {
6799         const struct bpf_sec_def *sec_def;
6800
6801         sec_def = find_sec_def(bpf_program__title(prog, false));
6802         if (!sec_def || !sec_def->attach_fn)
6803                 return ERR_PTR(-ESRCH);
6804
6805         return sec_def->attach_fn(sec_def, prog);
6806 }
6807
6808 enum bpf_perf_event_ret
6809 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
6810                            void **copy_mem, size_t *copy_size,
6811                            bpf_perf_event_print_t fn, void *private_data)
6812 {
6813         struct perf_event_mmap_page *header = mmap_mem;
6814         __u64 data_head = ring_buffer_read_head(header);
6815         __u64 data_tail = header->data_tail;
6816         void *base = ((__u8 *)header) + page_size;
6817         int ret = LIBBPF_PERF_EVENT_CONT;
6818         struct perf_event_header *ehdr;
6819         size_t ehdr_size;
6820
6821         while (data_head != data_tail) {
6822                 ehdr = base + (data_tail & (mmap_size - 1));
6823                 ehdr_size = ehdr->size;
6824
6825                 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
6826                         void *copy_start = ehdr;
6827                         size_t len_first = base + mmap_size - copy_start;
6828                         size_t len_secnd = ehdr_size - len_first;
6829
6830                         if (*copy_size < ehdr_size) {
6831                                 free(*copy_mem);
6832                                 *copy_mem = malloc(ehdr_size);
6833                                 if (!*copy_mem) {
6834                                         *copy_size = 0;
6835                                         ret = LIBBPF_PERF_EVENT_ERROR;
6836                                         break;
6837                                 }
6838                                 *copy_size = ehdr_size;
6839                         }
6840
6841                         memcpy(*copy_mem, copy_start, len_first);
6842                         memcpy(*copy_mem + len_first, base, len_secnd);
6843                         ehdr = *copy_mem;
6844                 }
6845
6846                 ret = fn(ehdr, private_data);
6847                 data_tail += ehdr_size;
6848                 if (ret != LIBBPF_PERF_EVENT_CONT)
6849                         break;
6850         }
6851
6852         ring_buffer_write_tail(header, data_tail);
6853         return ret;
6854 }
6855
6856 struct perf_buffer;
6857
6858 struct perf_buffer_params {
6859         struct perf_event_attr *attr;
6860         /* if event_cb is specified, it takes precendence */
6861         perf_buffer_event_fn event_cb;
6862         /* sample_cb and lost_cb are higher-level common-case callbacks */
6863         perf_buffer_sample_fn sample_cb;
6864         perf_buffer_lost_fn lost_cb;
6865         void *ctx;
6866         int cpu_cnt;
6867         int *cpus;
6868         int *map_keys;
6869 };
6870
6871 struct perf_cpu_buf {
6872         struct perf_buffer *pb;
6873         void *base; /* mmap()'ed memory */
6874         void *buf; /* for reconstructing segmented data */
6875         size_t buf_size;
6876         int fd;
6877         int cpu;
6878         int map_key;
6879 };
6880
6881 struct perf_buffer {
6882         perf_buffer_event_fn event_cb;
6883         perf_buffer_sample_fn sample_cb;
6884         perf_buffer_lost_fn lost_cb;
6885         void *ctx; /* passed into callbacks */
6886
6887         size_t page_size;
6888         size_t mmap_size;
6889         struct perf_cpu_buf **cpu_bufs;
6890         struct epoll_event *events;
6891         int cpu_cnt; /* number of allocated CPU buffers */
6892         int epoll_fd; /* perf event FD */
6893         int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
6894 };
6895
6896 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
6897                                       struct perf_cpu_buf *cpu_buf)
6898 {
6899         if (!cpu_buf)
6900                 return;
6901         if (cpu_buf->base &&
6902             munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
6903                 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
6904         if (cpu_buf->fd >= 0) {
6905                 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
6906                 close(cpu_buf->fd);
6907         }
6908         free(cpu_buf->buf);
6909         free(cpu_buf);
6910 }
6911
6912 void perf_buffer__free(struct perf_buffer *pb)
6913 {
6914         int i;
6915
6916         if (!pb)
6917                 return;
6918         if (pb->cpu_bufs) {
6919                 for (i = 0; i < pb->cpu_cnt && pb->cpu_bufs[i]; i++) {
6920                         struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
6921
6922                         bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
6923                         perf_buffer__free_cpu_buf(pb, cpu_buf);
6924                 }
6925                 free(pb->cpu_bufs);
6926         }
6927         if (pb->epoll_fd >= 0)
6928                 close(pb->epoll_fd);
6929         free(pb->events);
6930         free(pb);
6931 }
6932
6933 static struct perf_cpu_buf *
6934 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
6935                           int cpu, int map_key)
6936 {
6937         struct perf_cpu_buf *cpu_buf;
6938         char msg[STRERR_BUFSIZE];
6939         int err;
6940
6941         cpu_buf = calloc(1, sizeof(*cpu_buf));
6942         if (!cpu_buf)
6943                 return ERR_PTR(-ENOMEM);
6944
6945         cpu_buf->pb = pb;
6946         cpu_buf->cpu = cpu;
6947         cpu_buf->map_key = map_key;
6948
6949         cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
6950                               -1, PERF_FLAG_FD_CLOEXEC);
6951         if (cpu_buf->fd < 0) {
6952                 err = -errno;
6953                 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
6954                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
6955                 goto error;
6956         }
6957
6958         cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
6959                              PROT_READ | PROT_WRITE, MAP_SHARED,
6960                              cpu_buf->fd, 0);
6961         if (cpu_buf->base == MAP_FAILED) {
6962                 cpu_buf->base = NULL;
6963                 err = -errno;
6964                 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
6965                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
6966                 goto error;
6967         }
6968
6969         if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
6970                 err = -errno;
6971                 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
6972                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
6973                 goto error;
6974         }
6975
6976         return cpu_buf;
6977
6978 error:
6979         perf_buffer__free_cpu_buf(pb, cpu_buf);
6980         return (struct perf_cpu_buf *)ERR_PTR(err);
6981 }
6982
6983 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
6984                                               struct perf_buffer_params *p);
6985
6986 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
6987                                      const struct perf_buffer_opts *opts)
6988 {
6989         struct perf_buffer_params p = {};
6990         struct perf_event_attr attr = { 0, };
6991
6992         attr.config = PERF_COUNT_SW_BPF_OUTPUT,
6993         attr.type = PERF_TYPE_SOFTWARE;
6994         attr.sample_type = PERF_SAMPLE_RAW;
6995         attr.sample_period = 1;
6996         attr.wakeup_events = 1;
6997
6998         p.attr = &attr;
6999         p.sample_cb = opts ? opts->sample_cb : NULL;
7000         p.lost_cb = opts ? opts->lost_cb : NULL;
7001         p.ctx = opts ? opts->ctx : NULL;
7002
7003         return __perf_buffer__new(map_fd, page_cnt, &p);
7004 }
7005
7006 struct perf_buffer *
7007 perf_buffer__new_raw(int map_fd, size_t page_cnt,
7008                      const struct perf_buffer_raw_opts *opts)
7009 {
7010         struct perf_buffer_params p = {};
7011
7012         p.attr = opts->attr;
7013         p.event_cb = opts->event_cb;
7014         p.ctx = opts->ctx;
7015         p.cpu_cnt = opts->cpu_cnt;
7016         p.cpus = opts->cpus;
7017         p.map_keys = opts->map_keys;
7018
7019         return __perf_buffer__new(map_fd, page_cnt, &p);
7020 }
7021
7022 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
7023                                               struct perf_buffer_params *p)
7024 {
7025         const char *online_cpus_file = "/sys/devices/system/cpu/online";
7026         struct bpf_map_info map = {};
7027         char msg[STRERR_BUFSIZE];
7028         struct perf_buffer *pb;
7029         bool *online = NULL;
7030         __u32 map_info_len;
7031         int err, i, j, n;
7032
7033         if (page_cnt & (page_cnt - 1)) {
7034                 pr_warn("page count should be power of two, but is %zu\n",
7035                         page_cnt);
7036                 return ERR_PTR(-EINVAL);
7037         }
7038
7039         map_info_len = sizeof(map);
7040         err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
7041         if (err) {
7042                 err = -errno;
7043                 pr_warn("failed to get map info for map FD %d: %s\n",
7044                         map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
7045                 return ERR_PTR(err);
7046         }
7047
7048         if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
7049                 pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
7050                         map.name);
7051                 return ERR_PTR(-EINVAL);
7052         }
7053
7054         pb = calloc(1, sizeof(*pb));
7055         if (!pb)
7056                 return ERR_PTR(-ENOMEM);
7057
7058         pb->event_cb = p->event_cb;
7059         pb->sample_cb = p->sample_cb;
7060         pb->lost_cb = p->lost_cb;
7061         pb->ctx = p->ctx;
7062
7063         pb->page_size = getpagesize();
7064         pb->mmap_size = pb->page_size * page_cnt;
7065         pb->map_fd = map_fd;
7066
7067         pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
7068         if (pb->epoll_fd < 0) {
7069                 err = -errno;
7070                 pr_warn("failed to create epoll instance: %s\n",
7071                         libbpf_strerror_r(err, msg, sizeof(msg)));
7072                 goto error;
7073         }
7074
7075         if (p->cpu_cnt > 0) {
7076                 pb->cpu_cnt = p->cpu_cnt;
7077         } else {
7078                 pb->cpu_cnt = libbpf_num_possible_cpus();
7079                 if (pb->cpu_cnt < 0) {
7080                         err = pb->cpu_cnt;
7081                         goto error;
7082                 }
7083                 if (map.max_entries < pb->cpu_cnt)
7084                         pb->cpu_cnt = map.max_entries;
7085         }
7086
7087         pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
7088         if (!pb->events) {
7089                 err = -ENOMEM;
7090                 pr_warn("failed to allocate events: out of memory\n");
7091                 goto error;
7092         }
7093         pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
7094         if (!pb->cpu_bufs) {
7095                 err = -ENOMEM;
7096                 pr_warn("failed to allocate buffers: out of memory\n");
7097                 goto error;
7098         }
7099
7100         err = parse_cpu_mask_file(online_cpus_file, &online, &n);
7101         if (err) {
7102                 pr_warn("failed to get online CPU mask: %d\n", err);
7103                 goto error;
7104         }
7105
7106         for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
7107                 struct perf_cpu_buf *cpu_buf;
7108                 int cpu, map_key;
7109
7110                 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
7111                 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
7112
7113                 /* in case user didn't explicitly requested particular CPUs to
7114                  * be attached to, skip offline/not present CPUs
7115                  */
7116                 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
7117                         continue;
7118
7119                 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
7120                 if (IS_ERR(cpu_buf)) {
7121                         err = PTR_ERR(cpu_buf);
7122                         goto error;
7123                 }
7124
7125                 pb->cpu_bufs[j] = cpu_buf;
7126
7127                 err = bpf_map_update_elem(pb->map_fd, &map_key,
7128                                           &cpu_buf->fd, 0);
7129                 if (err) {
7130                         err = -errno;
7131                         pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
7132                                 cpu, map_key, cpu_buf->fd,
7133                                 libbpf_strerror_r(err, msg, sizeof(msg)));
7134                         goto error;
7135                 }
7136
7137                 pb->events[j].events = EPOLLIN;
7138                 pb->events[j].data.ptr = cpu_buf;
7139                 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
7140                               &pb->events[j]) < 0) {
7141                         err = -errno;
7142                         pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
7143                                 cpu, cpu_buf->fd,
7144                                 libbpf_strerror_r(err, msg, sizeof(msg)));
7145                         goto error;
7146                 }
7147                 j++;
7148         }
7149         pb->cpu_cnt = j;
7150         free(online);
7151
7152         return pb;
7153
7154 error:
7155         free(online);
7156         if (pb)
7157                 perf_buffer__free(pb);
7158         return ERR_PTR(err);
7159 }
7160
7161 struct perf_sample_raw {
7162         struct perf_event_header header;
7163         uint32_t size;
7164         char data[0];
7165 };
7166
7167 struct perf_sample_lost {
7168         struct perf_event_header header;
7169         uint64_t id;
7170         uint64_t lost;
7171         uint64_t sample_id;
7172 };
7173
7174 static enum bpf_perf_event_ret
7175 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
7176 {
7177         struct perf_cpu_buf *cpu_buf = ctx;
7178         struct perf_buffer *pb = cpu_buf->pb;
7179         void *data = e;
7180
7181         /* user wants full control over parsing perf event */
7182         if (pb->event_cb)
7183                 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
7184
7185         switch (e->type) {
7186         case PERF_RECORD_SAMPLE: {
7187                 struct perf_sample_raw *s = data;
7188
7189                 if (pb->sample_cb)
7190                         pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
7191                 break;
7192         }
7193         case PERF_RECORD_LOST: {
7194                 struct perf_sample_lost *s = data;
7195
7196                 if (pb->lost_cb)
7197                         pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
7198                 break;
7199         }
7200         default:
7201                 pr_warn("unknown perf sample type %d\n", e->type);
7202                 return LIBBPF_PERF_EVENT_ERROR;
7203         }
7204         return LIBBPF_PERF_EVENT_CONT;
7205 }
7206
7207 static int perf_buffer__process_records(struct perf_buffer *pb,
7208                                         struct perf_cpu_buf *cpu_buf)
7209 {
7210         enum bpf_perf_event_ret ret;
7211
7212         ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
7213                                          pb->page_size, &cpu_buf->buf,
7214                                          &cpu_buf->buf_size,
7215                                          perf_buffer__process_record, cpu_buf);
7216         if (ret != LIBBPF_PERF_EVENT_CONT)
7217                 return ret;
7218         return 0;
7219 }
7220
7221 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
7222 {
7223         int i, cnt, err;
7224
7225         cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
7226         for (i = 0; i < cnt; i++) {
7227                 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
7228
7229                 err = perf_buffer__process_records(pb, cpu_buf);
7230                 if (err) {
7231                         pr_warn("error while processing records: %d\n", err);
7232                         return err;
7233                 }
7234         }
7235         return cnt < 0 ? -errno : cnt;
7236 }
7237
7238 struct bpf_prog_info_array_desc {
7239         int     array_offset;   /* e.g. offset of jited_prog_insns */
7240         int     count_offset;   /* e.g. offset of jited_prog_len */
7241         int     size_offset;    /* > 0: offset of rec size,
7242                                  * < 0: fix size of -size_offset
7243                                  */
7244 };
7245
7246 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
7247         [BPF_PROG_INFO_JITED_INSNS] = {
7248                 offsetof(struct bpf_prog_info, jited_prog_insns),
7249                 offsetof(struct bpf_prog_info, jited_prog_len),
7250                 -1,
7251         },
7252         [BPF_PROG_INFO_XLATED_INSNS] = {
7253                 offsetof(struct bpf_prog_info, xlated_prog_insns),
7254                 offsetof(struct bpf_prog_info, xlated_prog_len),
7255                 -1,
7256         },
7257         [BPF_PROG_INFO_MAP_IDS] = {
7258                 offsetof(struct bpf_prog_info, map_ids),
7259                 offsetof(struct bpf_prog_info, nr_map_ids),
7260                 -(int)sizeof(__u32),
7261         },
7262         [BPF_PROG_INFO_JITED_KSYMS] = {
7263                 offsetof(struct bpf_prog_info, jited_ksyms),
7264                 offsetof(struct bpf_prog_info, nr_jited_ksyms),
7265                 -(int)sizeof(__u64),
7266         },
7267         [BPF_PROG_INFO_JITED_FUNC_LENS] = {
7268                 offsetof(struct bpf_prog_info, jited_func_lens),
7269                 offsetof(struct bpf_prog_info, nr_jited_func_lens),
7270                 -(int)sizeof(__u32),
7271         },
7272         [BPF_PROG_INFO_FUNC_INFO] = {
7273                 offsetof(struct bpf_prog_info, func_info),
7274                 offsetof(struct bpf_prog_info, nr_func_info),
7275                 offsetof(struct bpf_prog_info, func_info_rec_size),
7276         },
7277         [BPF_PROG_INFO_LINE_INFO] = {
7278                 offsetof(struct bpf_prog_info, line_info),
7279                 offsetof(struct bpf_prog_info, nr_line_info),
7280                 offsetof(struct bpf_prog_info, line_info_rec_size),
7281         },
7282         [BPF_PROG_INFO_JITED_LINE_INFO] = {
7283                 offsetof(struct bpf_prog_info, jited_line_info),
7284                 offsetof(struct bpf_prog_info, nr_jited_line_info),
7285                 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
7286         },
7287         [BPF_PROG_INFO_PROG_TAGS] = {
7288                 offsetof(struct bpf_prog_info, prog_tags),
7289                 offsetof(struct bpf_prog_info, nr_prog_tags),
7290                 -(int)sizeof(__u8) * BPF_TAG_SIZE,
7291         },
7292
7293 };
7294
7295 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info,
7296                                            int offset)
7297 {
7298         __u32 *array = (__u32 *)info;
7299
7300         if (offset >= 0)
7301                 return array[offset / sizeof(__u32)];
7302         return -(int)offset;
7303 }
7304
7305 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info,
7306                                            int offset)
7307 {
7308         __u64 *array = (__u64 *)info;
7309
7310         if (offset >= 0)
7311                 return array[offset / sizeof(__u64)];
7312         return -(int)offset;
7313 }
7314
7315 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
7316                                          __u32 val)
7317 {
7318         __u32 *array = (__u32 *)info;
7319
7320         if (offset >= 0)
7321                 array[offset / sizeof(__u32)] = val;
7322 }
7323
7324 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
7325                                          __u64 val)
7326 {
7327         __u64 *array = (__u64 *)info;
7328
7329         if (offset >= 0)
7330                 array[offset / sizeof(__u64)] = val;
7331 }
7332
7333 struct bpf_prog_info_linear *
7334 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
7335 {
7336         struct bpf_prog_info_linear *info_linear;
7337         struct bpf_prog_info info = {};
7338         __u32 info_len = sizeof(info);
7339         __u32 data_len = 0;
7340         int i, err;
7341         void *ptr;
7342
7343         if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
7344                 return ERR_PTR(-EINVAL);
7345
7346         /* step 1: get array dimensions */
7347         err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
7348         if (err) {
7349                 pr_debug("can't get prog info: %s", strerror(errno));
7350                 return ERR_PTR(-EFAULT);
7351         }
7352
7353         /* step 2: calculate total size of all arrays */
7354         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
7355                 bool include_array = (arrays & (1UL << i)) > 0;
7356                 struct bpf_prog_info_array_desc *desc;
7357                 __u32 count, size;
7358
7359                 desc = bpf_prog_info_array_desc + i;
7360
7361                 /* kernel is too old to support this field */
7362                 if (info_len < desc->array_offset + sizeof(__u32) ||
7363                     info_len < desc->count_offset + sizeof(__u32) ||
7364                     (desc->size_offset > 0 && info_len < desc->size_offset))
7365                         include_array = false;
7366
7367                 if (!include_array) {
7368                         arrays &= ~(1UL << i);  /* clear the bit */
7369                         continue;
7370                 }
7371
7372                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
7373                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
7374
7375                 data_len += count * size;
7376         }
7377
7378         /* step 3: allocate continuous memory */
7379         data_len = roundup(data_len, sizeof(__u64));
7380         info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
7381         if (!info_linear)
7382                 return ERR_PTR(-ENOMEM);
7383
7384         /* step 4: fill data to info_linear->info */
7385         info_linear->arrays = arrays;
7386         memset(&info_linear->info, 0, sizeof(info));
7387         ptr = info_linear->data;
7388
7389         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
7390                 struct bpf_prog_info_array_desc *desc;
7391                 __u32 count, size;
7392
7393                 if ((arrays & (1UL << i)) == 0)
7394                         continue;
7395
7396                 desc  = bpf_prog_info_array_desc + i;
7397                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
7398                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
7399                 bpf_prog_info_set_offset_u32(&info_linear->info,
7400                                              desc->count_offset, count);
7401                 bpf_prog_info_set_offset_u32(&info_linear->info,
7402                                              desc->size_offset, size);
7403                 bpf_prog_info_set_offset_u64(&info_linear->info,
7404                                              desc->array_offset,
7405                                              ptr_to_u64(ptr));
7406                 ptr += count * size;
7407         }
7408
7409         /* step 5: call syscall again to get required arrays */
7410         err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
7411         if (err) {
7412                 pr_debug("can't get prog info: %s", strerror(errno));
7413                 free(info_linear);
7414                 return ERR_PTR(-EFAULT);
7415         }
7416
7417         /* step 6: verify the data */
7418         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
7419                 struct bpf_prog_info_array_desc *desc;
7420                 __u32 v1, v2;
7421
7422                 if ((arrays & (1UL << i)) == 0)
7423                         continue;
7424
7425                 desc = bpf_prog_info_array_desc + i;
7426                 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
7427                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
7428                                                    desc->count_offset);
7429                 if (v1 != v2)
7430                         pr_warn("%s: mismatch in element count\n", __func__);
7431
7432                 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
7433                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
7434                                                    desc->size_offset);
7435                 if (v1 != v2)
7436                         pr_warn("%s: mismatch in rec size\n", __func__);
7437         }
7438
7439         /* step 7: update info_len and data_len */
7440         info_linear->info_len = sizeof(struct bpf_prog_info);
7441         info_linear->data_len = data_len;
7442
7443         return info_linear;
7444 }
7445
7446 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
7447 {
7448         int i;
7449
7450         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
7451                 struct bpf_prog_info_array_desc *desc;
7452                 __u64 addr, offs;
7453
7454                 if ((info_linear->arrays & (1UL << i)) == 0)
7455                         continue;
7456
7457                 desc = bpf_prog_info_array_desc + i;
7458                 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
7459                                                      desc->array_offset);
7460                 offs = addr - ptr_to_u64(info_linear->data);
7461                 bpf_prog_info_set_offset_u64(&info_linear->info,
7462                                              desc->array_offset, offs);
7463         }
7464 }
7465
7466 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
7467 {
7468         int i;
7469
7470         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
7471                 struct bpf_prog_info_array_desc *desc;
7472                 __u64 addr, offs;
7473
7474                 if ((info_linear->arrays & (1UL << i)) == 0)
7475                         continue;
7476
7477                 desc = bpf_prog_info_array_desc + i;
7478                 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
7479                                                      desc->array_offset);
7480                 addr = offs + ptr_to_u64(info_linear->data);
7481                 bpf_prog_info_set_offset_u64(&info_linear->info,
7482                                              desc->array_offset, addr);
7483         }
7484 }
7485
7486 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
7487 {
7488         int err = 0, n, len, start, end = -1;
7489         bool *tmp;
7490
7491         *mask = NULL;
7492         *mask_sz = 0;
7493
7494         /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
7495         while (*s) {
7496                 if (*s == ',' || *s == '\n') {
7497                         s++;
7498                         continue;
7499                 }
7500                 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
7501                 if (n <= 0 || n > 2) {
7502                         pr_warn("Failed to get CPU range %s: %d\n", s, n);
7503                         err = -EINVAL;
7504                         goto cleanup;
7505                 } else if (n == 1) {
7506                         end = start;
7507                 }
7508                 if (start < 0 || start > end) {
7509                         pr_warn("Invalid CPU range [%d,%d] in %s\n",
7510                                 start, end, s);
7511                         err = -EINVAL;
7512                         goto cleanup;
7513                 }
7514                 tmp = realloc(*mask, end + 1);
7515                 if (!tmp) {
7516                         err = -ENOMEM;
7517                         goto cleanup;
7518                 }
7519                 *mask = tmp;
7520                 memset(tmp + *mask_sz, 0, start - *mask_sz);
7521                 memset(tmp + start, 1, end - start + 1);
7522                 *mask_sz = end + 1;
7523                 s += len;
7524         }
7525         if (!*mask_sz) {
7526                 pr_warn("Empty CPU range\n");
7527                 return -EINVAL;
7528         }
7529         return 0;
7530 cleanup:
7531         free(*mask);
7532         *mask = NULL;
7533         return err;
7534 }
7535
7536 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
7537 {
7538         int fd, err = 0, len;
7539         char buf[128];
7540
7541         fd = open(fcpu, O_RDONLY);
7542         if (fd < 0) {
7543                 err = -errno;
7544                 pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err);
7545                 return err;
7546         }
7547         len = read(fd, buf, sizeof(buf));
7548         close(fd);
7549         if (len <= 0) {
7550                 err = len ? -errno : -EINVAL;
7551                 pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err);
7552                 return err;
7553         }
7554         if (len >= sizeof(buf)) {
7555                 pr_warn("CPU mask is too big in file %s\n", fcpu);
7556                 return -E2BIG;
7557         }
7558         buf[len] = '\0';
7559
7560         return parse_cpu_mask_str(buf, mask, mask_sz);
7561 }
7562
7563 int libbpf_num_possible_cpus(void)
7564 {
7565         static const char *fcpu = "/sys/devices/system/cpu/possible";
7566         static int cpus;
7567         int err, n, i, tmp_cpus;
7568         bool *mask;
7569
7570         tmp_cpus = READ_ONCE(cpus);
7571         if (tmp_cpus > 0)
7572                 return tmp_cpus;
7573
7574         err = parse_cpu_mask_file(fcpu, &mask, &n);
7575         if (err)
7576                 return err;
7577
7578         tmp_cpus = 0;
7579         for (i = 0; i < n; i++) {
7580                 if (mask[i])
7581                         tmp_cpus++;
7582         }
7583         free(mask);
7584
7585         WRITE_ONCE(cpus, tmp_cpus);
7586         return tmp_cpus;
7587 }
7588
7589 int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
7590                               const struct bpf_object_open_opts *opts)
7591 {
7592         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts,
7593                 .object_name = s->name,
7594         );
7595         struct bpf_object *obj;
7596         int i;
7597
7598         /* Attempt to preserve opts->object_name, unless overriden by user
7599          * explicitly. Overwriting object name for skeletons is discouraged,
7600          * as it breaks global data maps, because they contain object name
7601          * prefix as their own map name prefix. When skeleton is generated,
7602          * bpftool is making an assumption that this name will stay the same.
7603          */
7604         if (opts) {
7605                 memcpy(&skel_opts, opts, sizeof(*opts));
7606                 if (!opts->object_name)
7607                         skel_opts.object_name = s->name;
7608         }
7609
7610         obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts);
7611         if (IS_ERR(obj)) {
7612                 pr_warn("failed to initialize skeleton BPF object '%s': %ld\n",
7613                         s->name, PTR_ERR(obj));
7614                 return PTR_ERR(obj);
7615         }
7616
7617         *s->obj = obj;
7618
7619         for (i = 0; i < s->map_cnt; i++) {
7620                 struct bpf_map **map = s->maps[i].map;
7621                 const char *name = s->maps[i].name;
7622                 void **mmaped = s->maps[i].mmaped;
7623
7624                 *map = bpf_object__find_map_by_name(obj, name);
7625                 if (!*map) {
7626                         pr_warn("failed to find skeleton map '%s'\n", name);
7627                         return -ESRCH;
7628                 }
7629
7630                 /* externs shouldn't be pre-setup from user code */
7631                 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
7632                         *mmaped = (*map)->mmaped;
7633         }
7634
7635         for (i = 0; i < s->prog_cnt; i++) {
7636                 struct bpf_program **prog = s->progs[i].prog;
7637                 const char *name = s->progs[i].name;
7638
7639                 *prog = bpf_object__find_program_by_name(obj, name);
7640                 if (!*prog) {
7641                         pr_warn("failed to find skeleton program '%s'\n", name);
7642                         return -ESRCH;
7643                 }
7644         }
7645
7646         return 0;
7647 }
7648
7649 int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
7650 {
7651         int i, err;
7652
7653         err = bpf_object__load(*s->obj);
7654         if (err) {
7655                 pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err);
7656                 return err;
7657         }
7658
7659         for (i = 0; i < s->map_cnt; i++) {
7660                 struct bpf_map *map = *s->maps[i].map;
7661                 size_t mmap_sz = bpf_map_mmap_sz(map);
7662                 int prot, map_fd = bpf_map__fd(map);
7663                 void **mmaped = s->maps[i].mmaped;
7664
7665                 if (!mmaped)
7666                         continue;
7667
7668                 if (!(map->def.map_flags & BPF_F_MMAPABLE)) {
7669                         *mmaped = NULL;
7670                         continue;
7671                 }
7672
7673                 if (map->def.map_flags & BPF_F_RDONLY_PROG)
7674                         prot = PROT_READ;
7675                 else
7676                         prot = PROT_READ | PROT_WRITE;
7677
7678                 /* Remap anonymous mmap()-ed "map initialization image" as
7679                  * a BPF map-backed mmap()-ed memory, but preserving the same
7680                  * memory address. This will cause kernel to change process'
7681                  * page table to point to a different piece of kernel memory,
7682                  * but from userspace point of view memory address (and its
7683                  * contents, being identical at this point) will stay the
7684                  * same. This mapping will be released by bpf_object__close()
7685                  * as per normal clean up procedure, so we don't need to worry
7686                  * about it from skeleton's clean up perspective.
7687                  */
7688                 *mmaped = mmap(map->mmaped, mmap_sz, prot,
7689                                 MAP_SHARED | MAP_FIXED, map_fd, 0);
7690                 if (*mmaped == MAP_FAILED) {
7691                         err = -errno;
7692                         *mmaped = NULL;
7693                         pr_warn("failed to re-mmap() map '%s': %d\n",
7694                                  bpf_map__name(map), err);
7695                         return err;
7696                 }
7697         }
7698
7699         return 0;
7700 }
7701
7702 int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
7703 {
7704         int i;
7705
7706         for (i = 0; i < s->prog_cnt; i++) {
7707                 struct bpf_program *prog = *s->progs[i].prog;
7708                 struct bpf_link **link = s->progs[i].link;
7709                 const struct bpf_sec_def *sec_def;
7710                 const char *sec_name = bpf_program__title(prog, false);
7711
7712                 sec_def = find_sec_def(sec_name);
7713                 if (!sec_def || !sec_def->attach_fn)
7714                         continue;
7715
7716                 *link = sec_def->attach_fn(sec_def, prog);
7717                 if (IS_ERR(*link)) {
7718                         pr_warn("failed to auto-attach program '%s': %ld\n",
7719                                 bpf_program__name(prog), PTR_ERR(*link));
7720                         return PTR_ERR(*link);
7721                 }
7722         }
7723
7724         return 0;
7725 }
7726
7727 void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
7728 {
7729         int i;
7730
7731         for (i = 0; i < s->prog_cnt; i++) {
7732                 struct bpf_link **link = s->progs[i].link;
7733
7734                 if (!IS_ERR_OR_NULL(*link))
7735                         bpf_link__destroy(*link);
7736                 *link = NULL;
7737         }
7738 }
7739
7740 void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
7741 {
7742         if (s->progs)
7743                 bpf_object__detach_skeleton(s);
7744         if (s->obj)
7745                 bpf_object__close(*s->obj);
7746         free(s->maps);
7747         free(s->progs);
7748         free(s);
7749 }