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 <ctype.h>
  28 #include <asm/unistd.h>
  29 #include <linux/err.h>
  30 #include <linux/kernel.h>
  31 #include <linux/bpf.h>
  32 #include <linux/btf.h>
  33 #include <linux/filter.h>
  34 #include <linux/list.h>
  35 #include <linux/limits.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/ring_buffer.h>
  38 #include <linux/version.h>
  39 #include <sys/epoll.h>
  40 #include <sys/ioctl.h>
  41 #include <sys/mman.h>
  42 #include <sys/stat.h>
  43 #include <sys/types.h>
  44 #include <sys/vfs.h>
  45 #include <sys/utsname.h>
  46 #include <sys/resource.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 #define BPF_INSN_SZ (sizeof(struct bpf_insn))
  67
  68 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
  69  * compilation if user enables corresponding warning. Disable it explicitly.
  70  */
  71 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
  72
  73 #define __printf(a, b)  __attribute__((format(printf, a, b)))
  74
  75 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
  76 static const struct btf_type *
  77 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id);
  78
  79 static int __base_pr(enum libbpf_print_level level, const char *format,
  80                      va_list args)
  81 {
  82         if (level == LIBBPF_DEBUG)
  83                 return 0;
  84
  85         return vfprintf(stderr, format, args);
  86 }
  87
  88 static libbpf_print_fn_t __libbpf_pr = __base_pr;
  89
  90 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
  91 {
  92         libbpf_print_fn_t old_print_fn = __libbpf_pr;
  93
  94         __libbpf_pr = fn;
  95         return old_print_fn;
  96 }
  97
  98 __printf(2, 3)
  99 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
 100 {
 101         va_list args;
 102
 103         if (!__libbpf_pr)
 104                 return;
 105
 106         va_start(args, format);
 107         __libbpf_pr(level, format, args);
 108         va_end(args);
 109 }
 110
 111 static void pr_perm_msg(int err)
 112 {
 113         struct rlimit limit;
 114         char buf[100];
 115
 116         if (err != -EPERM || geteuid() != 0)
 117                 return;
 118
 119         err = getrlimit(RLIMIT_MEMLOCK, &limit);
 120         if (err)
 121                 return;
 122
 123         if (limit.rlim_cur == RLIM_INFINITY)
 124                 return;
 125
 126         if (limit.rlim_cur < 1024)
 127                 snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur);
 128         else if (limit.rlim_cur < 1024*1024)
 129                 snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024);
 130         else
 131                 snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024));
 132
 133         pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n",
 134                 buf);
 135 }
 136
 137 #define STRERR_BUFSIZE  128
 138
 139 /* Copied from tools/perf/util/util.h */
 140 #ifndef zfree
 141 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
 142 #endif
 143
 144 #ifndef zclose
 145 # define zclose(fd) ({                  \
 146         int ___err = 0;                 \
 147         if ((fd) >= 0)                  \
 148                 ___err = close((fd));   \
 149         fd = -1;                        \
 150         ___err; })
 151 #endif
 152
 153 static inline __u64 ptr_to_u64(const void *ptr)
 154 {
 155         return (__u64) (unsigned long) ptr;
 156 }
 157
 158 enum kern_feature_id {
 159         /* v4.14: kernel support for program & map names. */
 160         FEAT_PROG_NAME,
 161         /* v5.2: kernel support for global data sections. */
 162         FEAT_GLOBAL_DATA,
 163         /* BTF support */
 164         FEAT_BTF,
 165         /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
 166         FEAT_BTF_FUNC,
 167         /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
 168         FEAT_BTF_DATASEC,
 169         /* BTF_FUNC_GLOBAL is supported */
 170         FEAT_BTF_GLOBAL_FUNC,
 171         /* BPF_F_MMAPABLE is supported for arrays */
 172         FEAT_ARRAY_MMAP,
 173         /* kernel support for expected_attach_type in BPF_PROG_LOAD */
 174         FEAT_EXP_ATTACH_TYPE,
 175         /* bpf_probe_read_{kernel,user}[_str] helpers */
 176         FEAT_PROBE_READ_KERN,
 177         /* BPF_PROG_BIND_MAP is supported */
 178         FEAT_PROG_BIND_MAP,
 179         /* Kernel support for module BTFs */
 180         FEAT_MODULE_BTF,
 181         __FEAT_CNT,
 182 };
 183
 184 static bool kernel_supports(enum kern_feature_id feat_id);
 185
 186 enum reloc_type {
 187         RELO_LD64,
 188         RELO_CALL,
 189         RELO_DATA,
 190         RELO_EXTERN,
 191 };
 192
 193 struct reloc_desc {
 194         enum reloc_type type;
 195         int insn_idx;
 196         int map_idx;
 197         int sym_off;
 198         bool processed;
 199 };
 200
 201 struct bpf_sec_def;
 202
 203 typedef struct bpf_link *(*attach_fn_t)(const struct bpf_sec_def *sec,
 204                                         struct bpf_program *prog);
 205
 206 struct bpf_sec_def {
 207         const char *sec;
 208         size_t len;
 209         enum bpf_prog_type prog_type;
 210         enum bpf_attach_type expected_attach_type;
 211         bool is_exp_attach_type_optional;
 212         bool is_attachable;
 213         bool is_attach_btf;
 214         bool is_sleepable;
 215         attach_fn_t attach_fn;
 216 };
 217
 218 /*
 219  * bpf_prog should be a better name but it has been used in
 220  * linux/filter.h.
 221  */
 222 struct bpf_program {
 223         const struct bpf_sec_def *sec_def;
 224         char *sec_name;
 225         size_t sec_idx;
 226         /* this program's instruction offset (in number of instructions)
 227          * within its containing ELF section
 228          */
 229         size_t sec_insn_off;
 230         /* number of original instructions in ELF section belonging to this
 231          * program, not taking into account subprogram instructions possible
 232          * appended later during relocation
 233          */
 234         size_t sec_insn_cnt;
 235         /* Offset (in number of instructions) of the start of instruction
 236          * belonging to this BPF program  within its containing main BPF
 237          * program. For the entry-point (main) BPF program, this is always
 238          * zero. For a sub-program, this gets reset before each of main BPF
 239          * programs are processed and relocated and is used to determined
 240          * whether sub-program was already appended to the main program, and
 241          * if yes, at which instruction offset.
 242          */
 243         size_t sub_insn_off;
 244
 245         char *name;
 246         /* sec_name with / replaced by _; makes recursive pinning
 247          * in bpf_object__pin_programs easier
 248          */
 249         char *pin_name;
 250
 251         /* instructions that belong to BPF program; insns[0] is located at
 252          * sec_insn_off instruction within its ELF section in ELF file, so
 253          * when mapping ELF file instruction index to the local instruction,
 254          * one needs to subtract sec_insn_off; and vice versa.
 255          */
 256         struct bpf_insn *insns;
 257         /* actual number of instruction in this BPF program's image; for
 258          * entry-point BPF programs this includes the size of main program
 259          * itself plus all the used sub-programs, appended at the end
 260          */
 261         size_t insns_cnt;
 262
 263         struct reloc_desc *reloc_desc;
 264         int nr_reloc;
 265         int log_level;
 266
 267         struct {
 268                 int nr;
 269                 int *fds;
 270         } instances;
 271         bpf_program_prep_t preprocessor;
 272
 273         struct bpf_object *obj;
 274         void *priv;
 275         bpf_program_clear_priv_t clear_priv;
 276
 277         bool load;
 278         enum bpf_prog_type type;
 279         enum bpf_attach_type expected_attach_type;
 280         int prog_ifindex;
 281         __u32 attach_btf_obj_fd;
 282         __u32 attach_btf_id;
 283         __u32 attach_prog_fd;
 284         void *func_info;
 285         __u32 func_info_rec_size;
 286         __u32 func_info_cnt;
 287
 288         void *line_info;
 289         __u32 line_info_rec_size;
 290         __u32 line_info_cnt;
 291         __u32 prog_flags;
 292 };
 293
 294 struct bpf_struct_ops {
 295         const char *tname;
 296         const struct btf_type *type;
 297         struct bpf_program **progs;
 298         __u32 *kern_func_off;
 299         /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */
 300         void *data;
 301         /* e.g. struct bpf_struct_ops_tcp_congestion_ops in
 302          *      btf_vmlinux's format.
 303          * struct bpf_struct_ops_tcp_congestion_ops {
 304          *      [... some other kernel fields ...]
 305          *      struct tcp_congestion_ops data;
 306          * }
 307          * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops)
 308          * bpf_map__init_kern_struct_ops() will populate the "kern_vdata"
 309          * from "data".
 310          */
 311         void *kern_vdata;
 312         __u32 type_id;
 313 };
 314
 315 #define DATA_SEC ".data"
 316 #define BSS_SEC ".bss"
 317 #define RODATA_SEC ".rodata"
 318 #define KCONFIG_SEC ".kconfig"
 319 #define KSYMS_SEC ".ksyms"
 320 #define STRUCT_OPS_SEC ".struct_ops"
 321
 322 enum libbpf_map_type {
 323         LIBBPF_MAP_UNSPEC,
 324         LIBBPF_MAP_DATA,
 325         LIBBPF_MAP_BSS,
 326         LIBBPF_MAP_RODATA,
 327         LIBBPF_MAP_KCONFIG,
 328 };
 329
 330 static const char * const libbpf_type_to_btf_name[] = {
 331         [LIBBPF_MAP_DATA]       = DATA_SEC,
 332         [LIBBPF_MAP_BSS]        = BSS_SEC,
 333         [LIBBPF_MAP_RODATA]     = RODATA_SEC,
 334         [LIBBPF_MAP_KCONFIG]    = KCONFIG_SEC,
 335 };
 336
 337 struct bpf_map {
 338         char *name;
 339         int fd;
 340         int sec_idx;
 341         size_t sec_offset;
 342         int map_ifindex;
 343         int inner_map_fd;
 344         struct bpf_map_def def;
 345         __u32 numa_node;
 346         __u32 btf_var_idx;
 347         __u32 btf_key_type_id;
 348         __u32 btf_value_type_id;
 349         __u32 btf_vmlinux_value_type_id;
 350         void *priv;
 351         bpf_map_clear_priv_t clear_priv;
 352         enum libbpf_map_type libbpf_type;
 353         void *mmaped;
 354         struct bpf_struct_ops *st_ops;
 355         struct bpf_map *inner_map;
 356         void **init_slots;
 357         int init_slots_sz;
 358         char *pin_path;
 359         bool pinned;
 360         bool reused;
 361 };
 362
 363 enum extern_type {
 364         EXT_UNKNOWN,
 365         EXT_KCFG,
 366         EXT_KSYM,
 367 };
 368
 369 enum kcfg_type {
 370         KCFG_UNKNOWN,
 371         KCFG_CHAR,
 372         KCFG_BOOL,
 373         KCFG_INT,
 374         KCFG_TRISTATE,
 375         KCFG_CHAR_ARR,
 376 };
 377
 378 struct extern_desc {
 379         enum extern_type type;
 380         int sym_idx;
 381         int btf_id;
 382         int sec_btf_id;
 383         const char *name;
 384         bool is_set;
 385         bool is_weak;
 386         union {
 387                 struct {
 388                         enum kcfg_type type;
 389                         int sz;
 390                         int align;
 391                         int data_off;
 392                         bool is_signed;
 393                 } kcfg;
 394                 struct {
 395                         unsigned long long addr;
 396
 397                         /* target btf_id of the corresponding kernel var. */
 398                         int kernel_btf_obj_fd;
 399                         int kernel_btf_id;
 400
 401                         /* local btf_id of the ksym extern's type. */
 402                         __u32 type_id;
 403                 } ksym;
 404         };
 405 };
 406
 407 static LIST_HEAD(bpf_objects_list);
 408
 409 struct module_btf {
 410         struct btf *btf;
 411         char *name;
 412         __u32 id;
 413         int fd;
 414 };
 415
 416 struct bpf_object {
 417         char name[BPF_OBJ_NAME_LEN];
 418         char license[64];
 419         __u32 kern_version;
 420
 421         struct bpf_program *programs;
 422         size_t nr_programs;
 423         struct bpf_map *maps;
 424         size_t nr_maps;
 425         size_t maps_cap;
 426
 427         char *kconfig;
 428         struct extern_desc *externs;
 429         int nr_extern;
 430         int kconfig_map_idx;
 431         int rodata_map_idx;
 432
 433         bool loaded;
 434         bool has_subcalls;
 435
 436         /*
 437          * Information when doing elf related work. Only valid if fd
 438          * is valid.
 439          */
 440         struct {
 441                 int fd;
 442                 const void *obj_buf;
 443                 size_t obj_buf_sz;
 444                 Elf *elf;
 445                 GElf_Ehdr ehdr;
 446                 Elf_Data *symbols;
 447                 Elf_Data *data;
 448                 Elf_Data *rodata;
 449                 Elf_Data *bss;
 450                 Elf_Data *st_ops_data;
 451                 size_t shstrndx; /* section index for section name strings */
 452                 size_t strtabidx;
 453                 struct {
 454                         GElf_Shdr shdr;
 455                         Elf_Data *data;
 456                 } *reloc_sects;
 457                 int nr_reloc_sects;
 458                 int maps_shndx;
 459                 int btf_maps_shndx;
 460                 __u32 btf_maps_sec_btf_id;
 461                 int text_shndx;
 462                 int symbols_shndx;
 463                 int data_shndx;
 464                 int rodata_shndx;
 465                 int bss_shndx;
 466                 int st_ops_shndx;
 467         } efile;
 468         /*
 469          * All loaded bpf_object is linked in a list, which is
 470          * hidden to caller. bpf_objects__<func> handlers deal with
 471          * all objects.
 472          */
 473         struct list_head list;
 474
 475         struct btf *btf;
 476         struct btf_ext *btf_ext;
 477
 478         /* Parse and load BTF vmlinux if any of the programs in the object need
 479          * it at load time.
 480          */
 481         struct btf *btf_vmlinux;
 482         /* vmlinux BTF override for CO-RE relocations */
 483         struct btf *btf_vmlinux_override;
 484         /* Lazily initialized kernel module BTFs */
 485         struct module_btf *btf_modules;
 486         bool btf_modules_loaded;
 487         size_t btf_module_cnt;
 488         size_t btf_module_cap;
 489
 490         void *priv;
 491         bpf_object_clear_priv_t clear_priv;
 492
 493         char path[];
 494 };
 495 #define obj_elf_valid(o)        ((o)->efile.elf)
 496
 497 static const char *elf_sym_str(const struct bpf_object *obj, size_t off);
 498 static const char *elf_sec_str(const struct bpf_object *obj, size_t off);
 499 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx);
 500 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name);
 501 static int elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn, GElf_Shdr *hdr);
 502 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn);
 503 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn);
 504 static int elf_sym_by_sec_off(const struct bpf_object *obj, size_t sec_idx,
 505                               size_t off, __u32 sym_type, GElf_Sym *sym);
 506
 507 void bpf_program__unload(struct bpf_program *prog)
 508 {
 509         int i;
 510
 511         if (!prog)
 512                 return;
 513
 514         /*
 515          * If the object is opened but the program was never loaded,
 516          * it is possible that prog->instances.nr == -1.
 517          */
 518         if (prog->instances.nr > 0) {
 519                 for (i = 0; i < prog->instances.nr; i++)
 520                         zclose(prog->instances.fds[i]);
 521         } else if (prog->instances.nr != -1) {
 522                 pr_warn("Internal error: instances.nr is %d\n",
 523                         prog->instances.nr);
 524         }
 525
 526         prog->instances.nr = -1;
 527         zfree(&prog->instances.fds);
 528
 529         zfree(&prog->func_info);
 530         zfree(&prog->line_info);
 531 }
 532
 533 static void bpf_program__exit(struct bpf_program *prog)
 534 {
 535         if (!prog)
 536                 return;
 537
 538         if (prog->clear_priv)
 539                 prog->clear_priv(prog, prog->priv);
 540
 541         prog->priv = NULL;
 542         prog->clear_priv = NULL;
 543
 544         bpf_program__unload(prog);
 545         zfree(&prog->name);
 546         zfree(&prog->sec_name);
 547         zfree(&prog->pin_name);
 548         zfree(&prog->insns);
 549         zfree(&prog->reloc_desc);
 550
 551         prog->nr_reloc = 0;
 552         prog->insns_cnt = 0;
 553         prog->sec_idx = -1;
 554 }
 555
 556 static char *__bpf_program__pin_name(struct bpf_program *prog)
 557 {
 558         char *name, *p;
 559
 560         name = p = strdup(prog->sec_name);
 561         while ((p = strchr(p, '/')))
 562                 *p = '_';
 563
 564         return name;
 565 }
 566
 567 static bool insn_is_subprog_call(const struct bpf_insn *insn)
 568 {
 569         return BPF_CLASS(insn->code) == BPF_JMP &&
 570                BPF_OP(insn->code) == BPF_CALL &&
 571                BPF_SRC(insn->code) == BPF_K &&
 572                insn->src_reg == BPF_PSEUDO_CALL &&
 573                insn->dst_reg == 0 &&
 574                insn->off == 0;
 575 }
 576
 577 static int
 578 bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog,
 579                       const char *name, size_t sec_idx, const char *sec_name,
 580                       size_t sec_off, void *insn_data, size_t insn_data_sz)
 581 {
 582         if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) {
 583                 pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n",
 584                         sec_name, name, sec_off, insn_data_sz);
 585                 return -EINVAL;
 586         }
 587
 588         memset(prog, 0, sizeof(*prog));
 589         prog->obj = obj;
 590
 591         prog->sec_idx = sec_idx;
 592         prog->sec_insn_off = sec_off / BPF_INSN_SZ;
 593         prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ;
 594         /* insns_cnt can later be increased by appending used subprograms */
 595         prog->insns_cnt = prog->sec_insn_cnt;
 596
 597         prog->type = BPF_PROG_TYPE_UNSPEC;
 598         prog->load = true;
 599
 600         prog->instances.fds = NULL;
 601         prog->instances.nr = -1;
 602
 603         prog->sec_name = strdup(sec_name);
 604         if (!prog->sec_name)
 605                 goto errout;
 606
 607         prog->name = strdup(name);
 608         if (!prog->name)
 609                 goto errout;
 610
 611         prog->pin_name = __bpf_program__pin_name(prog);
 612         if (!prog->pin_name)
 613                 goto errout;
 614
 615         prog->insns = malloc(insn_data_sz);
 616         if (!prog->insns)
 617                 goto errout;
 618         memcpy(prog->insns, insn_data, insn_data_sz);
 619
 620         return 0;
 621 errout:
 622         pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name);
 623         bpf_program__exit(prog);
 624         return -ENOMEM;
 625 }
 626
 627 static int
 628 bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data,
 629                          const char *sec_name, int sec_idx)
 630 {
 631         struct bpf_program *prog, *progs;
 632         void *data = sec_data->d_buf;
 633         size_t sec_sz = sec_data->d_size, sec_off, prog_sz;
 634         int nr_progs, err;
 635         const char *name;
 636         GElf_Sym sym;
 637
 638         progs = obj->programs;
 639         nr_progs = obj->nr_programs;
 640         sec_off = 0;
 641
 642         while (sec_off < sec_sz) {
 643                 if (elf_sym_by_sec_off(obj, sec_idx, sec_off, STT_FUNC, &sym)) {
 644                         pr_warn("sec '%s': failed to find program symbol at offset %zu\n",
 645                                 sec_name, sec_off);
 646                         return -LIBBPF_ERRNO__FORMAT;
 647                 }
 648
 649                 prog_sz = sym.st_size;
 650
 651                 name = elf_sym_str(obj, sym.st_name);
 652                 if (!name) {
 653                         pr_warn("sec '%s': failed to get symbol name for offset %zu\n",
 654                                 sec_name, sec_off);
 655                         return -LIBBPF_ERRNO__FORMAT;
 656                 }
 657
 658                 if (sec_off + prog_sz > sec_sz) {
 659                         pr_warn("sec '%s': program at offset %zu crosses section boundary\n",
 660                                 sec_name, sec_off);
 661                         return -LIBBPF_ERRNO__FORMAT;
 662                 }
 663
 664                 pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n",
 665                          sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz);
 666
 667                 progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs));
 668                 if (!progs) {
 669                         /*
 670                          * In this case the original obj->programs
 671                          * is still valid, so don't need special treat for
 672                          * bpf_close_object().
 673                          */
 674                         pr_warn("sec '%s': failed to alloc memory for new program '%s'\n",
 675                                 sec_name, name);
 676                         return -ENOMEM;
 677                 }
 678                 obj->programs = progs;
 679
 680                 prog = &progs[nr_progs];
 681
 682                 err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name,
 683                                             sec_off, data + sec_off, prog_sz);
 684                 if (err)
 685                         return err;
 686
 687                 nr_progs++;
 688                 obj->nr_programs = nr_progs;
 689
 690                 sec_off += prog_sz;
 691         }
 692
 693         return 0;
 694 }
 695
 696 static __u32 get_kernel_version(void)
 697 {
 698         __u32 major, minor, patch;
 699         struct utsname info;
 700
 701         uname(&info);
 702         if (sscanf(info.release, "%u.%u.%u", &major, &minor, &patch) != 3)
 703                 return 0;
 704         return KERNEL_VERSION(major, minor, patch);
 705 }
 706
 707 static const struct btf_member *
 708 find_member_by_offset(const struct btf_type *t, __u32 bit_offset)
 709 {
 710         struct btf_member *m;
 711         int i;
 712
 713         for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
 714                 if (btf_member_bit_offset(t, i) == bit_offset)
 715                         return m;
 716         }
 717
 718         return NULL;
 719 }
 720
 721 static const struct btf_member *
 722 find_member_by_name(const struct btf *btf, const struct btf_type *t,
 723                     const char *name)
 724 {
 725         struct btf_member *m;
 726         int i;
 727
 728         for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
 729                 if (!strcmp(btf__name_by_offset(btf, m->name_off), name))
 730                         return m;
 731         }
 732
 733         return NULL;
 734 }
 735
 736 #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_"
 737 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
 738                                    const char *name, __u32 kind);
 739
 740 static int
 741 find_struct_ops_kern_types(const struct btf *btf, const char *tname,
 742                            const struct btf_type **type, __u32 *type_id,
 743                            const struct btf_type **vtype, __u32 *vtype_id,
 744                            const struct btf_member **data_member)
 745 {
 746         const struct btf_type *kern_type, *kern_vtype;
 747         const struct btf_member *kern_data_member;
 748         __s32 kern_vtype_id, kern_type_id;
 749         __u32 i;
 750
 751         kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
 752         if (kern_type_id < 0) {
 753                 pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n",
 754                         tname);
 755                 return kern_type_id;
 756         }
 757         kern_type = btf__type_by_id(btf, kern_type_id);
 758
 759         /* Find the corresponding "map_value" type that will be used
 760          * in map_update(BPF_MAP_TYPE_STRUCT_OPS).  For example,
 761          * find "struct bpf_struct_ops_tcp_congestion_ops" from the
 762          * btf_vmlinux.
 763          */
 764         kern_vtype_id = find_btf_by_prefix_kind(btf, STRUCT_OPS_VALUE_PREFIX,
 765                                                 tname, BTF_KIND_STRUCT);
 766         if (kern_vtype_id < 0) {
 767                 pr_warn("struct_ops init_kern: struct %s%s is not found in kernel BTF\n",
 768                         STRUCT_OPS_VALUE_PREFIX, tname);
 769                 return kern_vtype_id;
 770         }
 771         kern_vtype = btf__type_by_id(btf, kern_vtype_id);
 772
 773         /* Find "struct tcp_congestion_ops" from
 774          * struct bpf_struct_ops_tcp_congestion_ops {
 775          *      [ ... ]
 776          *      struct tcp_congestion_ops data;
 777          * }
 778          */
 779         kern_data_member = btf_members(kern_vtype);
 780         for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) {
 781                 if (kern_data_member->type == kern_type_id)
 782                         break;
 783         }
 784         if (i == btf_vlen(kern_vtype)) {
 785                 pr_warn("struct_ops init_kern: struct %s data is not found in struct %s%s\n",
 786                         tname, STRUCT_OPS_VALUE_PREFIX, tname);
 787                 return -EINVAL;
 788         }
 789
 790         *type = kern_type;
 791         *type_id = kern_type_id;
 792         *vtype = kern_vtype;
 793         *vtype_id = kern_vtype_id;
 794         *data_member = kern_data_member;
 795
 796         return 0;
 797 }
 798
 799 static bool bpf_map__is_struct_ops(const struct bpf_map *map)
 800 {
 801         return map->def.type == BPF_MAP_TYPE_STRUCT_OPS;
 802 }
 803
 804 /* Init the map's fields that depend on kern_btf */
 805 static int bpf_map__init_kern_struct_ops(struct bpf_map *map,
 806                                          const struct btf *btf,
 807                                          const struct btf *kern_btf)
 808 {
 809         const struct btf_member *member, *kern_member, *kern_data_member;
 810         const struct btf_type *type, *kern_type, *kern_vtype;
 811         __u32 i, kern_type_id, kern_vtype_id, kern_data_off;
 812         struct bpf_struct_ops *st_ops;
 813         void *data, *kern_data;
 814         const char *tname;
 815         int err;
 816
 817         st_ops = map->st_ops;
 818         type = st_ops->type;
 819         tname = st_ops->tname;
 820         err = find_struct_ops_kern_types(kern_btf, tname,
 821                                          &kern_type, &kern_type_id,
 822                                          &kern_vtype, &kern_vtype_id,
 823                                          &kern_data_member);
 824         if (err)
 825                 return err;
 826
 827         pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n",
 828                  map->name, st_ops->type_id, kern_type_id, kern_vtype_id);
 829
 830         map->def.value_size = kern_vtype->size;
 831         map->btf_vmlinux_value_type_id = kern_vtype_id;
 832
 833         st_ops->kern_vdata = calloc(1, kern_vtype->size);
 834         if (!st_ops->kern_vdata)
 835                 return -ENOMEM;
 836
 837         data = st_ops->data;
 838         kern_data_off = kern_data_member->offset / 8;
 839         kern_data = st_ops->kern_vdata + kern_data_off;
 840
 841         member = btf_members(type);
 842         for (i = 0; i < btf_vlen(type); i++, member++) {
 843                 const struct btf_type *mtype, *kern_mtype;
 844                 __u32 mtype_id, kern_mtype_id;
 845                 void *mdata, *kern_mdata;
 846                 __s64 msize, kern_msize;
 847                 __u32 moff, kern_moff;
 848                 __u32 kern_member_idx;
 849                 const char *mname;
 850
 851                 mname = btf__name_by_offset(btf, member->name_off);
 852                 kern_member = find_member_by_name(kern_btf, kern_type, mname);
 853                 if (!kern_member) {
 854                         pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n",
 855                                 map->name, mname);
 856                         return -ENOTSUP;
 857                 }
 858
 859                 kern_member_idx = kern_member - btf_members(kern_type);
 860                 if (btf_member_bitfield_size(type, i) ||
 861                     btf_member_bitfield_size(kern_type, kern_member_idx)) {
 862                         pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n",
 863                                 map->name, mname);
 864                         return -ENOTSUP;
 865                 }
 866
 867                 moff = member->offset / 8;
 868                 kern_moff = kern_member->offset / 8;
 869
 870                 mdata = data + moff;
 871                 kern_mdata = kern_data + kern_moff;
 872
 873                 mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id);
 874                 kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type,
 875                                                     &kern_mtype_id);
 876                 if (BTF_INFO_KIND(mtype->info) !=
 877                     BTF_INFO_KIND(kern_mtype->info)) {
 878                         pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n",
 879                                 map->name, mname, BTF_INFO_KIND(mtype->info),
 880                                 BTF_INFO_KIND(kern_mtype->info));
 881                         return -ENOTSUP;
 882                 }
 883
 884                 if (btf_is_ptr(mtype)) {
 885                         struct bpf_program *prog;
 886
 887                         mtype = skip_mods_and_typedefs(btf, mtype->type, &mtype_id);
 888                         kern_mtype = skip_mods_and_typedefs(kern_btf,
 889                                                             kern_mtype->type,
 890                                                             &kern_mtype_id);
 891                         if (!btf_is_func_proto(mtype) ||
 892                             !btf_is_func_proto(kern_mtype)) {
 893                                 pr_warn("struct_ops init_kern %s: non func ptr %s is not supported\n",
 894                                         map->name, mname);
 895                                 return -ENOTSUP;
 896                         }
 897
 898                         prog = st_ops->progs[i];
 899                         if (!prog) {
 900                                 pr_debug("struct_ops init_kern %s: func ptr %s is not set\n",
 901                                          map->name, mname);
 902                                 continue;
 903                         }
 904
 905                         prog->attach_btf_id = kern_type_id;
 906                         prog->expected_attach_type = kern_member_idx;
 907
 908                         st_ops->kern_func_off[i] = kern_data_off + kern_moff;
 909
 910                         pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n",
 911                                  map->name, mname, prog->name, moff,
 912                                  kern_moff);
 913
 914                         continue;
 915                 }
 916
 917                 msize = btf__resolve_size(btf, mtype_id);
 918                 kern_msize = btf__resolve_size(kern_btf, kern_mtype_id);
 919                 if (msize < 0 || kern_msize < 0 || msize != kern_msize) {
 920                         pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n",
 921                                 map->name, mname, (ssize_t)msize,
 922                                 (ssize_t)kern_msize);
 923                         return -ENOTSUP;
 924                 }
 925
 926                 pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n",
 927                          map->name, mname, (unsigned int)msize,
 928                          moff, kern_moff);
 929                 memcpy(kern_mdata, mdata, msize);
 930         }
 931
 932         return 0;
 933 }
 934
 935 static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj)
 936 {
 937         struct bpf_map *map;
 938         size_t i;
 939         int err;
 940
 941         for (i = 0; i < obj->nr_maps; i++) {
 942                 map = &obj->maps[i];
 943
 944                 if (!bpf_map__is_struct_ops(map))
 945                         continue;
 946
 947                 err = bpf_map__init_kern_struct_ops(map, obj->btf,
 948                                                     obj->btf_vmlinux);
 949                 if (err)
 950                         return err;
 951         }
 952
 953         return 0;
 954 }
 955
 956 static int bpf_object__init_struct_ops_maps(struct bpf_object *obj)
 957 {
 958         const struct btf_type *type, *datasec;
 959         const struct btf_var_secinfo *vsi;
 960         struct bpf_struct_ops *st_ops;
 961         const char *tname, *var_name;
 962         __s32 type_id, datasec_id;
 963         const struct btf *btf;
 964         struct bpf_map *map;
 965         __u32 i;
 966
 967         if (obj->efile.st_ops_shndx == -1)
 968                 return 0;
 969
 970         btf = obj->btf;
 971         datasec_id = btf__find_by_name_kind(btf, STRUCT_OPS_SEC,
 972                                             BTF_KIND_DATASEC);
 973         if (datasec_id < 0) {
 974                 pr_warn("struct_ops init: DATASEC %s not found\n",
 975                         STRUCT_OPS_SEC);
 976                 return -EINVAL;
 977         }
 978
 979         datasec = btf__type_by_id(btf, datasec_id);
 980         vsi = btf_var_secinfos(datasec);
 981         for (i = 0; i < btf_vlen(datasec); i++, vsi++) {
 982                 type = btf__type_by_id(obj->btf, vsi->type);
 983                 var_name = btf__name_by_offset(obj->btf, type->name_off);
 984
 985                 type_id = btf__resolve_type(obj->btf, vsi->type);
 986                 if (type_id < 0) {
 987                         pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n",
 988                                 vsi->type, STRUCT_OPS_SEC);
 989                         return -EINVAL;
 990                 }
 991
 992                 type = btf__type_by_id(obj->btf, type_id);
 993                 tname = btf__name_by_offset(obj->btf, type->name_off);
 994                 if (!tname[0]) {
 995                         pr_warn("struct_ops init: anonymous type is not supported\n");
 996                         return -ENOTSUP;
 997                 }
 998                 if (!btf_is_struct(type)) {
 999                         pr_warn("struct_ops init: %s is not a struct\n", tname);
1000                         return -EINVAL;
1001                 }
1002
1003                 map = bpf_object__add_map(obj);
1004                 if (IS_ERR(map))
1005                         return PTR_ERR(map);
1006
1007                 map->sec_idx = obj->efile.st_ops_shndx;
1008                 map->sec_offset = vsi->offset;
1009                 map->name = strdup(var_name);
1010                 if (!map->name)
1011                         return -ENOMEM;
1012
1013                 map->def.type = BPF_MAP_TYPE_STRUCT_OPS;
1014                 map->def.key_size = sizeof(int);
1015                 map->def.value_size = type->size;
1016                 map->def.max_entries = 1;
1017
1018                 map->st_ops = calloc(1, sizeof(*map->st_ops));
1019                 if (!map->st_ops)
1020                         return -ENOMEM;
1021                 st_ops = map->st_ops;
1022                 st_ops->data = malloc(type->size);
1023                 st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs));
1024                 st_ops->kern_func_off = malloc(btf_vlen(type) *
1025                                                sizeof(*st_ops->kern_func_off));
1026                 if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off)
1027                         return -ENOMEM;
1028
1029                 if (vsi->offset + type->size > obj->efile.st_ops_data->d_size) {
1030                         pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n",
1031                                 var_name, STRUCT_OPS_SEC);
1032                         return -EINVAL;
1033                 }
1034
1035                 memcpy(st_ops->data,
1036                        obj->efile.st_ops_data->d_buf + vsi->offset,
1037                        type->size);
1038                 st_ops->tname = tname;
1039                 st_ops->type = type;
1040                 st_ops->type_id = type_id;
1041
1042                 pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n",
1043                          tname, type_id, var_name, vsi->offset);
1044         }
1045
1046         return 0;
1047 }
1048
1049 static struct bpf_object *bpf_object__new(const char *path,
1050                                           const void *obj_buf,
1051                                           size_t obj_buf_sz,
1052                                           const char *obj_name)
1053 {
1054         struct bpf_object *obj;
1055         char *end;
1056
1057         obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
1058         if (!obj) {
1059                 pr_warn("alloc memory failed for %s\n", path);
1060                 return ERR_PTR(-ENOMEM);
1061         }
1062
1063         strcpy(obj->path, path);
1064         if (obj_name) {
1065                 strncpy(obj->name, obj_name, sizeof(obj->name) - 1);
1066                 obj->name[sizeof(obj->name) - 1] = 0;
1067         } else {
1068                 /* Using basename() GNU version which doesn't modify arg. */
1069                 strncpy(obj->name, basename((void *)path),
1070                         sizeof(obj->name) - 1);
1071                 end = strchr(obj->name, '.');
1072                 if (end)
1073                         *end = 0;
1074         }
1075
1076         obj->efile.fd = -1;
1077         /*
1078          * Caller of this function should also call
1079          * bpf_object__elf_finish() after data collection to return
1080          * obj_buf to user. If not, we should duplicate the buffer to
1081          * avoid user freeing them before elf finish.
1082          */
1083         obj->efile.obj_buf = obj_buf;
1084         obj->efile.obj_buf_sz = obj_buf_sz;
1085         obj->efile.maps_shndx = -1;
1086         obj->efile.btf_maps_shndx = -1;
1087         obj->efile.data_shndx = -1;
1088         obj->efile.rodata_shndx = -1;
1089         obj->efile.bss_shndx = -1;
1090         obj->efile.st_ops_shndx = -1;
1091         obj->kconfig_map_idx = -1;
1092         obj->rodata_map_idx = -1;
1093
1094         obj->kern_version = get_kernel_version();
1095         obj->loaded = false;
1096
1097         INIT_LIST_HEAD(&obj->list);
1098         list_add(&obj->list, &bpf_objects_list);
1099         return obj;
1100 }
1101
1102 static void bpf_object__elf_finish(struct bpf_object *obj)
1103 {
1104         if (!obj_elf_valid(obj))
1105                 return;
1106
1107         if (obj->efile.elf) {
1108                 elf_end(obj->efile.elf);
1109                 obj->efile.elf = NULL;
1110         }
1111         obj->efile.symbols = NULL;
1112         obj->efile.data = NULL;
1113         obj->efile.rodata = NULL;
1114         obj->efile.bss = NULL;
1115         obj->efile.st_ops_data = NULL;
1116
1117         zfree(&obj->efile.reloc_sects);
1118         obj->efile.nr_reloc_sects = 0;
1119         zclose(obj->efile.fd);
1120         obj->efile.obj_buf = NULL;
1121         obj->efile.obj_buf_sz = 0;
1122 }
1123
1124 /* if libelf is old and doesn't support mmap(), fall back to read() */
1125 #ifndef ELF_C_READ_MMAP
1126 #define ELF_C_READ_MMAP ELF_C_READ
1127 #endif
1128
1129 static int bpf_object__elf_init(struct bpf_object *obj)
1130 {
1131         int err = 0;
1132         GElf_Ehdr *ep;
1133
1134         if (obj_elf_valid(obj)) {
1135                 pr_warn("elf: init internal error\n");
1136                 return -LIBBPF_ERRNO__LIBELF;
1137         }
1138
1139         if (obj->efile.obj_buf_sz > 0) {
1140                 /*
1141                  * obj_buf should have been validated by
1142                  * bpf_object__open_buffer().
1143                  */
1144                 obj->efile.elf = elf_memory((char *)obj->efile.obj_buf,
1145                                             obj->efile.obj_buf_sz);
1146         } else {
1147                 obj->efile.fd = open(obj->path, O_RDONLY);
1148                 if (obj->efile.fd < 0) {
1149                         char errmsg[STRERR_BUFSIZE], *cp;
1150
1151                         err = -errno;
1152                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
1153                         pr_warn("elf: failed to open %s: %s\n", obj->path, cp);
1154                         return err;
1155                 }
1156
1157                 obj->efile.elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL);
1158         }
1159
1160         if (!obj->efile.elf) {
1161                 pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1));
1162                 err = -LIBBPF_ERRNO__LIBELF;
1163                 goto errout;
1164         }
1165
1166         if (!gelf_getehdr(obj->efile.elf, &obj->efile.ehdr)) {
1167                 pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1));
1168                 err = -LIBBPF_ERRNO__FORMAT;
1169                 goto errout;
1170         }
1171         ep = &obj->efile.ehdr;
1172
1173         if (elf_getshdrstrndx(obj->efile.elf, &obj->efile.shstrndx)) {
1174                 pr_warn("elf: failed to get section names section index for %s: %s\n",
1175                         obj->path, elf_errmsg(-1));
1176                 err = -LIBBPF_ERRNO__FORMAT;
1177                 goto errout;
1178         }
1179
1180         /* Elf is corrupted/truncated, avoid calling elf_strptr. */
1181         if (!elf_rawdata(elf_getscn(obj->efile.elf, obj->efile.shstrndx), NULL)) {
1182                 pr_warn("elf: failed to get section names strings from %s: %s\n",
1183                         obj->path, elf_errmsg(-1));
1184                 return -LIBBPF_ERRNO__FORMAT;
1185         }
1186
1187         /* Old LLVM set e_machine to EM_NONE */
1188         if (ep->e_type != ET_REL ||
1189             (ep->e_machine && ep->e_machine != EM_BPF)) {
1190                 pr_warn("elf: %s is not a valid eBPF object file\n", obj->path);
1191                 err = -LIBBPF_ERRNO__FORMAT;
1192                 goto errout;
1193         }
1194
1195         return 0;
1196 errout:
1197         bpf_object__elf_finish(obj);
1198         return err;
1199 }
1200
1201 static int bpf_object__check_endianness(struct bpf_object *obj)
1202 {
1203 #if __BYTE_ORDER == __LITTLE_ENDIAN
1204         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2LSB)
1205                 return 0;
1206 #elif __BYTE_ORDER == __BIG_ENDIAN
1207         if (obj->efile.ehdr.e_ident[EI_DATA] == ELFDATA2MSB)
1208                 return 0;
1209 #else
1210 # error "Unrecognized __BYTE_ORDER__"
1211 #endif
1212         pr_warn("elf: endianness mismatch in %s.\n", obj->path);
1213         return -LIBBPF_ERRNO__ENDIAN;
1214 }
1215
1216 static int
1217 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
1218 {
1219         memcpy(obj->license, data, min(size, sizeof(obj->license) - 1));
1220         pr_debug("license of %s is %s\n", obj->path, obj->license);
1221         return 0;
1222 }
1223
1224 static int
1225 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
1226 {
1227         __u32 kver;
1228
1229         if (size != sizeof(kver)) {
1230                 pr_warn("invalid kver section in %s\n", obj->path);
1231                 return -LIBBPF_ERRNO__FORMAT;
1232         }
1233         memcpy(&kver, data, sizeof(kver));
1234         obj->kern_version = kver;
1235         pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
1236         return 0;
1237 }
1238
1239 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
1240 {
1241         if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1242             type == BPF_MAP_TYPE_HASH_OF_MAPS)
1243                 return true;
1244         return false;
1245 }
1246
1247 int bpf_object__section_size(const struct bpf_object *obj, const char *name,
1248                              __u32 *size)
1249 {
1250         int ret = -ENOENT;
1251
1252         *size = 0;
1253         if (!name) {
1254                 return -EINVAL;
1255         } else if (!strcmp(name, DATA_SEC)) {
1256                 if (obj->efile.data)
1257                         *size = obj->efile.data->d_size;
1258         } else if (!strcmp(name, BSS_SEC)) {
1259                 if (obj->efile.bss)
1260                         *size = obj->efile.bss->d_size;
1261         } else if (!strcmp(name, RODATA_SEC)) {
1262                 if (obj->efile.rodata)
1263                         *size = obj->efile.rodata->d_size;
1264         } else if (!strcmp(name, STRUCT_OPS_SEC)) {
1265                 if (obj->efile.st_ops_data)
1266                         *size = obj->efile.st_ops_data->d_size;
1267         } else {
1268                 Elf_Scn *scn = elf_sec_by_name(obj, name);
1269                 Elf_Data *data = elf_sec_data(obj, scn);
1270
1271                 if (data) {
1272                         ret = 0; /* found it */
1273                         *size = data->d_size;
1274                 }
1275         }
1276
1277         return *size ? 0 : ret;
1278 }
1279
1280 int bpf_object__variable_offset(const struct bpf_object *obj, const char *name,
1281                                 __u32 *off)
1282 {
1283         Elf_Data *symbols = obj->efile.symbols;
1284         const char *sname;
1285         size_t si;
1286
1287         if (!name || !off)
1288                 return -EINVAL;
1289
1290         for (si = 0; si < symbols->d_size / sizeof(GElf_Sym); si++) {
1291                 GElf_Sym sym;
1292
1293                 if (!gelf_getsym(symbols, si, &sym))
1294                         continue;
1295                 if (GELF_ST_BIND(sym.st_info) != STB_GLOBAL ||
1296                     GELF_ST_TYPE(sym.st_info) != STT_OBJECT)
1297                         continue;
1298
1299                 sname = elf_sym_str(obj, sym.st_name);
1300                 if (!sname) {
1301                         pr_warn("failed to get sym name string for var %s\n",
1302                                 name);
1303                         return -EIO;
1304                 }
1305                 if (strcmp(name, sname) == 0) {
1306                         *off = sym.st_value;
1307                         return 0;
1308                 }
1309         }
1310
1311         return -ENOENT;
1312 }
1313
1314 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
1315 {
1316         struct bpf_map *new_maps;
1317         size_t new_cap;
1318         int i;
1319
1320         if (obj->nr_maps < obj->maps_cap)
1321                 return &obj->maps[obj->nr_maps++];
1322
1323         new_cap = max((size_t)4, obj->maps_cap * 3 / 2);
1324         new_maps = libbpf_reallocarray(obj->maps, new_cap, sizeof(*obj->maps));
1325         if (!new_maps) {
1326                 pr_warn("alloc maps for object failed\n");
1327                 return ERR_PTR(-ENOMEM);
1328         }
1329
1330         obj->maps_cap = new_cap;
1331         obj->maps = new_maps;
1332
1333         /* zero out new maps */
1334         memset(obj->maps + obj->nr_maps, 0,
1335                (obj->maps_cap - obj->nr_maps) * sizeof(*obj->maps));
1336         /*
1337          * fill all fd with -1 so won't close incorrect fd (fd=0 is stdin)
1338          * when failure (zclose won't close negative fd)).
1339          */
1340         for (i = obj->nr_maps; i < obj->maps_cap; i++) {
1341                 obj->maps[i].fd = -1;
1342                 obj->maps[i].inner_map_fd = -1;
1343         }
1344
1345         return &obj->maps[obj->nr_maps++];
1346 }
1347
1348 static size_t bpf_map_mmap_sz(const struct bpf_map *map)
1349 {
1350         long page_sz = sysconf(_SC_PAGE_SIZE);
1351         size_t map_sz;
1352
1353         map_sz = (size_t)roundup(map->def.value_size, 8) * map->def.max_entries;
1354         map_sz = roundup(map_sz, page_sz);
1355         return map_sz;
1356 }
1357
1358 static char *internal_map_name(struct bpf_object *obj,
1359                                enum libbpf_map_type type)
1360 {
1361         char map_name[BPF_OBJ_NAME_LEN], *p;
1362         const char *sfx = libbpf_type_to_btf_name[type];
1363         int sfx_len = max((size_t)7, strlen(sfx));
1364         int pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1,
1365                           strlen(obj->name));
1366
1367         snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
1368                  sfx_len, libbpf_type_to_btf_name[type]);
1369
1370         /* sanitise map name to characters allowed by kernel */
1371         for (p = map_name; *p && p < map_name + sizeof(map_name); p++)
1372                 if (!isalnum(*p) && *p != '_' && *p != '.')
1373                         *p = '_';
1374
1375         return strdup(map_name);
1376 }
1377
1378 static int
1379 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
1380                               int sec_idx, void *data, size_t data_sz)
1381 {
1382         struct bpf_map_def *def;
1383         struct bpf_map *map;
1384         int err;
1385
1386         map = bpf_object__add_map(obj);
1387         if (IS_ERR(map))
1388                 return PTR_ERR(map);
1389
1390         map->libbpf_type = type;
1391         map->sec_idx = sec_idx;
1392         map->sec_offset = 0;
1393         map->name = internal_map_name(obj, type);
1394         if (!map->name) {
1395                 pr_warn("failed to alloc map name\n");
1396                 return -ENOMEM;
1397         }
1398
1399         def = &map->def;
1400         def->type = BPF_MAP_TYPE_ARRAY;
1401         def->key_size = sizeof(int);
1402         def->value_size = data_sz;
1403         def->max_entries = 1;
1404         def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
1405                          ? BPF_F_RDONLY_PROG : 0;
1406         def->map_flags |= BPF_F_MMAPABLE;
1407
1408         pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
1409                  map->name, map->sec_idx, map->sec_offset, def->map_flags);
1410
1411         map->mmaped = mmap(NULL, bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE,
1412                            MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1413         if (map->mmaped == MAP_FAILED) {
1414                 err = -errno;
1415                 map->mmaped = NULL;
1416                 pr_warn("failed to alloc map '%s' content buffer: %d\n",
1417                         map->name, err);
1418                 zfree(&map->name);
1419                 return err;
1420         }
1421
1422         if (data)
1423                 memcpy(map->mmaped, data, data_sz);
1424
1425         pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
1426         return 0;
1427 }
1428
1429 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
1430 {
1431         int err;
1432
1433         /*
1434          * Populate obj->maps with libbpf internal maps.
1435          */
1436         if (obj->efile.data_shndx >= 0) {
1437                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
1438                                                     obj->efile.data_shndx,
1439                                                     obj->efile.data->d_buf,
1440                                                     obj->efile.data->d_size);
1441                 if (err)
1442                         return err;
1443         }
1444         if (obj->efile.rodata_shndx >= 0) {
1445                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
1446                                                     obj->efile.rodata_shndx,
1447                                                     obj->efile.rodata->d_buf,
1448                                                     obj->efile.rodata->d_size);
1449                 if (err)
1450                         return err;
1451
1452                 obj->rodata_map_idx = obj->nr_maps - 1;
1453         }
1454         if (obj->efile.bss_shndx >= 0) {
1455                 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
1456                                                     obj->efile.bss_shndx,
1457                                                     NULL,
1458                                                     obj->efile.bss->d_size);
1459                 if (err)
1460                         return err;
1461         }
1462         return 0;
1463 }
1464
1465
1466 static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
1467                                                const void *name)
1468 {
1469         int i;
1470
1471         for (i = 0; i < obj->nr_extern; i++) {
1472                 if (strcmp(obj->externs[i].name, name) == 0)
1473                         return &obj->externs[i];
1474         }
1475         return NULL;
1476 }
1477
1478 static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val,
1479                               char value)
1480 {
1481         switch (ext->kcfg.type) {
1482         case KCFG_BOOL:
1483                 if (value == 'm') {
1484                         pr_warn("extern (kcfg) %s=%c should be tristate or char\n",
1485                                 ext->name, value);
1486                         return -EINVAL;
1487                 }
1488                 *(bool *)ext_val = value == 'y' ? true : false;
1489                 break;
1490         case KCFG_TRISTATE:
1491                 if (value == 'y')
1492                         *(enum libbpf_tristate *)ext_val = TRI_YES;
1493                 else if (value == 'm')
1494                         *(enum libbpf_tristate *)ext_val = TRI_MODULE;
1495                 else /* value == 'n' */
1496                         *(enum libbpf_tristate *)ext_val = TRI_NO;
1497                 break;
1498         case KCFG_CHAR:
1499                 *(char *)ext_val = value;
1500                 break;
1501         case KCFG_UNKNOWN:
1502         case KCFG_INT:
1503         case KCFG_CHAR_ARR:
1504         default:
1505                 pr_warn("extern (kcfg) %s=%c should be bool, tristate, or char\n",
1506                         ext->name, value);
1507                 return -EINVAL;
1508         }
1509         ext->is_set = true;
1510         return 0;
1511 }
1512
1513 static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val,
1514                               const char *value)
1515 {
1516         size_t len;
1517
1518         if (ext->kcfg.type != KCFG_CHAR_ARR) {
1519                 pr_warn("extern (kcfg) %s=%s should be char array\n", ext->name, value);
1520                 return -EINVAL;
1521         }
1522
1523         len = strlen(value);
1524         if (value[len - 1] != '"') {
1525                 pr_warn("extern (kcfg) '%s': invalid string config '%s'\n",
1526                         ext->name, value);
1527                 return -EINVAL;
1528         }
1529
1530         /* strip quotes */
1531         len -= 2;
1532         if (len >= ext->kcfg.sz) {
1533                 pr_warn("extern (kcfg) '%s': long string config %s of (%zu bytes) truncated to %d bytes\n",
1534                         ext->name, value, len, ext->kcfg.sz - 1);
1535                 len = ext->kcfg.sz - 1;
1536         }
1537         memcpy(ext_val, value + 1, len);
1538         ext_val[len] = '\0';
1539         ext->is_set = true;
1540         return 0;
1541 }
1542
1543 static int parse_u64(const char *value, __u64 *res)
1544 {
1545         char *value_end;
1546         int err;
1547
1548         errno = 0;
1549         *res = strtoull(value, &value_end, 0);
1550         if (errno) {
1551                 err = -errno;
1552                 pr_warn("failed to parse '%s' as integer: %d\n", value, err);
1553                 return err;
1554         }
1555         if (*value_end) {
1556                 pr_warn("failed to parse '%s' as integer completely\n", value);
1557                 return -EINVAL;
1558         }
1559         return 0;
1560 }
1561
1562 static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v)
1563 {
1564         int bit_sz = ext->kcfg.sz * 8;
1565
1566         if (ext->kcfg.sz == 8)
1567                 return true;
1568
1569         /* Validate that value stored in u64 fits in integer of `ext->sz`
1570          * bytes size without any loss of information. If the target integer
1571          * is signed, we rely on the following limits of integer type of
1572          * Y bits and subsequent transformation:
1573          *
1574          *     -2^(Y-1) <= X           <= 2^(Y-1) - 1
1575          *            0 <= X + 2^(Y-1) <= 2^Y - 1
1576          *            0 <= X + 2^(Y-1) <  2^Y
1577          *
1578          *  For unsigned target integer, check that all the (64 - Y) bits are
1579          *  zero.
1580          */
1581         if (ext->kcfg.is_signed)
1582                 return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
1583         else
1584                 return (v >> bit_sz) == 0;
1585 }
1586
1587 static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val,
1588                               __u64 value)
1589 {
1590         if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) {
1591                 pr_warn("extern (kcfg) %s=%llu should be integer\n",
1592                         ext->name, (unsigned long long)value);
1593                 return -EINVAL;
1594         }
1595         if (!is_kcfg_value_in_range(ext, value)) {
1596                 pr_warn("extern (kcfg) %s=%llu value doesn't fit in %d bytes\n",
1597                         ext->name, (unsigned long long)value, ext->kcfg.sz);
1598                 return -ERANGE;
1599         }
1600         switch (ext->kcfg.sz) {
1601                 case 1: *(__u8 *)ext_val = value; break;
1602                 case 2: *(__u16 *)ext_val = value; break;
1603                 case 4: *(__u32 *)ext_val = value; break;
1604                 case 8: *(__u64 *)ext_val = value; break;
1605                 default:
1606                         return -EINVAL;
1607         }
1608         ext->is_set = true;
1609         return 0;
1610 }
1611
1612 static int bpf_object__process_kconfig_line(struct bpf_object *obj,
1613                                             char *buf, void *data)
1614 {
1615         struct extern_desc *ext;
1616         char *sep, *value;
1617         int len, err = 0;
1618         void *ext_val;
1619         __u64 num;
1620
1621         if (strncmp(buf, "CONFIG_", 7))
1622                 return 0;
1623
1624         sep = strchr(buf, '=');
1625         if (!sep) {
1626                 pr_warn("failed to parse '%s': no separator\n", buf);
1627                 return -EINVAL;
1628         }
1629
1630         /* Trim ending '\n' */
1631         len = strlen(buf);
1632         if (buf[len - 1] == '\n')
1633                 buf[len - 1] = '\0';
1634         /* Split on '=' and ensure that a value is present. */
1635         *sep = '\0';
1636         if (!sep[1]) {
1637                 *sep = '=';
1638                 pr_warn("failed to parse '%s': no value\n", buf);
1639                 return -EINVAL;
1640         }
1641
1642         ext = find_extern_by_name(obj, buf);
1643         if (!ext || ext->is_set)
1644                 return 0;
1645
1646         ext_val = data + ext->kcfg.data_off;
1647         value = sep + 1;
1648
1649         switch (*value) {
1650         case 'y': case 'n': case 'm':
1651                 err = set_kcfg_value_tri(ext, ext_val, *value);
1652                 break;
1653         case '"':
1654                 err = set_kcfg_value_str(ext, ext_val, value);
1655                 break;
1656         default:
1657                 /* assume integer */
1658                 err = parse_u64(value, &num);
1659                 if (err) {
1660                         pr_warn("extern (kcfg) %s=%s should be integer\n",
1661                                 ext->name, value);
1662                         return err;
1663                 }
1664                 err = set_kcfg_value_num(ext, ext_val, num);
1665                 break;
1666         }
1667         if (err)
1668                 return err;
1669         pr_debug("extern (kcfg) %s=%s\n", ext->name, value);
1670         return 0;
1671 }
1672
1673 static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
1674 {
1675         char buf[PATH_MAX];
1676         struct utsname uts;
1677         int len, err = 0;
1678         gzFile file;
1679
1680         uname(&uts);
1681         len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
1682         if (len < 0)
1683                 return -EINVAL;
1684         else if (len >= PATH_MAX)
1685                 return -ENAMETOOLONG;
1686
1687         /* gzopen also accepts uncompressed files. */
1688         file = gzopen(buf, "r");
1689         if (!file)
1690                 file = gzopen("/proc/config.gz", "r");
1691
1692         if (!file) {
1693                 pr_warn("failed to open system Kconfig\n");
1694                 return -ENOENT;
1695         }
1696
1697         while (gzgets(file, buf, sizeof(buf))) {
1698                 err = bpf_object__process_kconfig_line(obj, buf, data);
1699                 if (err) {
1700                         pr_warn("error parsing system Kconfig line '%s': %d\n",
1701                                 buf, err);
1702                         goto out;
1703                 }
1704         }
1705
1706 out:
1707         gzclose(file);
1708         return err;
1709 }
1710
1711 static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
1712                                         const char *config, void *data)
1713 {
1714         char buf[PATH_MAX];
1715         int err = 0;
1716         FILE *file;
1717
1718         file = fmemopen((void *)config, strlen(config), "r");
1719         if (!file) {
1720                 err = -errno;
1721                 pr_warn("failed to open in-memory Kconfig: %d\n", err);
1722                 return err;
1723         }
1724
1725         while (fgets(buf, sizeof(buf), file)) {
1726                 err = bpf_object__process_kconfig_line(obj, buf, data);
1727                 if (err) {
1728                         pr_warn("error parsing in-memory Kconfig line '%s': %d\n",
1729                                 buf, err);
1730                         break;
1731                 }
1732         }
1733
1734         fclose(file);
1735         return err;
1736 }
1737
1738 static int bpf_object__init_kconfig_map(struct bpf_object *obj)
1739 {
1740         struct extern_desc *last_ext = NULL, *ext;
1741         size_t map_sz;
1742         int i, err;
1743
1744         for (i = 0; i < obj->nr_extern; i++) {
1745                 ext = &obj->externs[i];
1746                 if (ext->type == EXT_KCFG)
1747                         last_ext = ext;
1748         }
1749
1750         if (!last_ext)
1751                 return 0;
1752
1753         map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz;
1754         err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
1755                                             obj->efile.symbols_shndx,
1756                                             NULL, map_sz);
1757         if (err)
1758                 return err;
1759
1760         obj->kconfig_map_idx = obj->nr_maps - 1;
1761
1762         return 0;
1763 }
1764
1765 static int bpf_object__init_user_maps(struct bpf_object *obj, bool strict)
1766 {
1767         Elf_Data *symbols = obj->efile.symbols;
1768         int i, map_def_sz = 0, nr_maps = 0, nr_syms;
1769         Elf_Data *data = NULL;
1770         Elf_Scn *scn;
1771
1772         if (obj->efile.maps_shndx < 0)
1773                 return 0;
1774
1775         if (!symbols)
1776                 return -EINVAL;
1777
1778
1779         scn = elf_sec_by_idx(obj, obj->efile.maps_shndx);
1780         data = elf_sec_data(obj, scn);
1781         if (!scn || !data) {
1782                 pr_warn("elf: failed to get legacy map definitions for %s\n",
1783                         obj->path);
1784                 return -EINVAL;
1785         }
1786
1787         /*
1788          * Count number of maps. Each map has a name.
1789          * Array of maps is not supported: only the first element is
1790          * considered.
1791          *
1792          * TODO: Detect array of map and report error.
1793          */
1794         nr_syms = symbols->d_size / sizeof(GElf_Sym);
1795         for (i = 0; i < nr_syms; i++) {
1796                 GElf_Sym sym;
1797
1798                 if (!gelf_getsym(symbols, i, &sym))
1799                         continue;
1800                 if (sym.st_shndx != obj->efile.maps_shndx)
1801                         continue;
1802                 nr_maps++;
1803         }
1804         /* Assume equally sized map definitions */
1805         pr_debug("elf: found %d legacy map definitions (%zd bytes) in %s\n",
1806                  nr_maps, data->d_size, obj->path);
1807
1808         if (!data->d_size || nr_maps == 0 || (data->d_size % nr_maps) != 0) {
1809                 pr_warn("elf: unable to determine legacy map definition size in %s\n",
1810                         obj->path);
1811                 return -EINVAL;
1812         }
1813         map_def_sz = data->d_size / nr_maps;
1814
1815         /* Fill obj->maps using data in "maps" section.  */
1816         for (i = 0; i < nr_syms; i++) {
1817                 GElf_Sym sym;
1818                 const char *map_name;
1819                 struct bpf_map_def *def;
1820                 struct bpf_map *map;
1821
1822                 if (!gelf_getsym(symbols, i, &sym))
1823                         continue;
1824                 if (sym.st_shndx != obj->efile.maps_shndx)
1825                         continue;
1826
1827                 map = bpf_object__add_map(obj);
1828                 if (IS_ERR(map))
1829                         return PTR_ERR(map);
1830
1831                 map_name = elf_sym_str(obj, sym.st_name);
1832                 if (!map_name) {
1833                         pr_warn("failed to get map #%d name sym string for obj %s\n",
1834                                 i, obj->path);
1835                         return -LIBBPF_ERRNO__FORMAT;
1836                 }
1837
1838                 map->libbpf_type = LIBBPF_MAP_UNSPEC;
1839                 map->sec_idx = sym.st_shndx;
1840                 map->sec_offset = sym.st_value;
1841                 pr_debug("map '%s' (legacy): at sec_idx %d, offset %zu.\n",
1842                          map_name, map->sec_idx, map->sec_offset);
1843                 if (sym.st_value + map_def_sz > data->d_size) {
1844                         pr_warn("corrupted maps section in %s: last map \"%s\" too small\n",
1845                                 obj->path, map_name);
1846                         return -EINVAL;
1847                 }
1848
1849                 map->name = strdup(map_name);
1850                 if (!map->name) {
1851                         pr_warn("failed to alloc map name\n");
1852                         return -ENOMEM;
1853                 }
1854                 pr_debug("map %d is \"%s\"\n", i, map->name);
1855                 def = (struct bpf_map_def *)(data->d_buf + sym.st_value);
1856                 /*
1857                  * If the definition of the map in the object file fits in
1858                  * bpf_map_def, copy it.  Any extra fields in our version
1859                  * of bpf_map_def will default to zero as a result of the
1860                  * calloc above.
1861                  */
1862                 if (map_def_sz <= sizeof(struct bpf_map_def)) {
1863                         memcpy(&map->def, def, map_def_sz);
1864                 } else {
1865                         /*
1866                          * Here the map structure being read is bigger than what
1867                          * we expect, truncate if the excess bits are all zero.
1868                          * If they are not zero, reject this map as
1869                          * incompatible.
1870                          */
1871                         char *b;
1872
1873                         for (b = ((char *)def) + sizeof(struct bpf_map_def);
1874                              b < ((char *)def) + map_def_sz; b++) {
1875                                 if (*b != 0) {
1876                                         pr_warn("maps section in %s: \"%s\" has unrecognized, non-zero options\n",
1877                                                 obj->path, map_name);
1878                                         if (strict)
1879                                                 return -EINVAL;
1880                                 }
1881                         }
1882                         memcpy(&map->def, def, sizeof(struct bpf_map_def));
1883                 }
1884         }
1885         return 0;
1886 }
1887
1888 static const struct btf_type *
1889 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
1890 {
1891         const struct btf_type *t = btf__type_by_id(btf, id);
1892
1893         if (res_id)
1894                 *res_id = id;
1895
1896         while (btf_is_mod(t) || btf_is_typedef(t)) {
1897                 if (res_id)
1898                         *res_id = t->type;
1899                 t = btf__type_by_id(btf, t->type);
1900         }
1901
1902         return t;
1903 }
1904
1905 static const struct btf_type *
1906 resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id)
1907 {
1908         const struct btf_type *t;
1909
1910         t = skip_mods_and_typedefs(btf, id, NULL);
1911         if (!btf_is_ptr(t))
1912                 return NULL;
1913
1914         t = skip_mods_and_typedefs(btf, t->type, res_id);
1915
1916         return btf_is_func_proto(t) ? t : NULL;
1917 }
1918
1919 static const char *btf_kind_str(const struct btf_type *t)
1920 {
1921         switch (btf_kind(t)) {
1922         case BTF_KIND_UNKN: return "void";
1923         case BTF_KIND_INT: return "int";
1924         case BTF_KIND_PTR: return "ptr";
1925         case BTF_KIND_ARRAY: return "array";
1926         case BTF_KIND_STRUCT: return "struct";
1927         case BTF_KIND_UNION: return "union";
1928         case BTF_KIND_ENUM: return "enum";
1929         case BTF_KIND_FWD: return "fwd";
1930         case BTF_KIND_TYPEDEF: return "typedef";
1931         case BTF_KIND_VOLATILE: return "volatile";
1932         case BTF_KIND_CONST: return "const";
1933         case BTF_KIND_RESTRICT: return "restrict";
1934         case BTF_KIND_FUNC: return "func";
1935         case BTF_KIND_FUNC_PROTO: return "func_proto";
1936         case BTF_KIND_VAR: return "var";
1937         case BTF_KIND_DATASEC: return "datasec";
1938         default: return "unknown";
1939         }
1940 }
1941
1942 /*
1943  * Fetch integer attribute of BTF map definition. Such attributes are
1944  * represented using a pointer to an array, in which dimensionality of array
1945  * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
1946  * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
1947  * type definition, while using only sizeof(void *) space in ELF data section.
1948  */
1949 static bool get_map_field_int(const char *map_name, const struct btf *btf,
1950                               const struct btf_member *m, __u32 *res)
1951 {
1952         const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
1953         const char *name = btf__name_by_offset(btf, m->name_off);
1954         const struct btf_array *arr_info;
1955         const struct btf_type *arr_t;
1956
1957         if (!btf_is_ptr(t)) {
1958                 pr_warn("map '%s': attr '%s': expected PTR, got %s.\n",
1959                         map_name, name, btf_kind_str(t));
1960                 return false;
1961         }
1962
1963         arr_t = btf__type_by_id(btf, t->type);
1964         if (!arr_t) {
1965                 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
1966                         map_name, name, t->type);
1967                 return false;
1968         }
1969         if (!btf_is_array(arr_t)) {
1970                 pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n",
1971                         map_name, name, btf_kind_str(arr_t));
1972                 return false;
1973         }
1974         arr_info = btf_array(arr_t);
1975         *res = arr_info->nelems;
1976         return true;
1977 }
1978
1979 static int build_map_pin_path(struct bpf_map *map, const char *path)
1980 {
1981         char buf[PATH_MAX];
1982         int len;
1983
1984         if (!path)
1985                 path = "/sys/fs/bpf";
1986
1987         len = snprintf(buf, PATH_MAX, "%s/%s", path, bpf_map__name(map));
1988         if (len < 0)
1989                 return -EINVAL;
1990         else if (len >= PATH_MAX)
1991                 return -ENAMETOOLONG;
1992
1993         return bpf_map__set_pin_path(map, buf);
1994 }
1995
1996
1997 static int parse_btf_map_def(struct bpf_object *obj,
1998                              struct bpf_map *map,
1999                              const struct btf_type *def,
2000                              bool strict, bool is_inner,
2001                              const char *pin_root_path)
2002 {
2003         const struct btf_type *t;
2004         const struct btf_member *m;
2005         int vlen, i;
2006
2007         vlen = btf_vlen(def);
2008         m = btf_members(def);
2009         for (i = 0; i < vlen; i++, m++) {
2010                 const char *name = btf__name_by_offset(obj->btf, m->name_off);
2011
2012                 if (!name) {
2013                         pr_warn("map '%s': invalid field #%d.\n", map->name, i);
2014                         return -EINVAL;
2015                 }
2016                 if (strcmp(name, "type") == 0) {
2017                         if (!get_map_field_int(map->name, obj->btf, m,
2018                                                &map->def.type))
2019                                 return -EINVAL;
2020                         pr_debug("map '%s': found type = %u.\n",
2021                                  map->name, map->def.type);
2022                 } else if (strcmp(name, "max_entries") == 0) {
2023                         if (!get_map_field_int(map->name, obj->btf, m,
2024                                                &map->def.max_entries))
2025                                 return -EINVAL;
2026                         pr_debug("map '%s': found max_entries = %u.\n",
2027                                  map->name, map->def.max_entries);
2028                 } else if (strcmp(name, "map_flags") == 0) {
2029                         if (!get_map_field_int(map->name, obj->btf, m,
2030                                                &map->def.map_flags))
2031                                 return -EINVAL;
2032                         pr_debug("map '%s': found map_flags = %u.\n",
2033                                  map->name, map->def.map_flags);
2034                 } else if (strcmp(name, "numa_node") == 0) {
2035                         if (!get_map_field_int(map->name, obj->btf, m, &map->numa_node))
2036                                 return -EINVAL;
2037                         pr_debug("map '%s': found numa_node = %u.\n", map->name, map->numa_node);
2038                 } else if (strcmp(name, "key_size") == 0) {
2039                         __u32 sz;
2040
2041                         if (!get_map_field_int(map->name, obj->btf, m, &sz))
2042                                 return -EINVAL;
2043                         pr_debug("map '%s': found key_size = %u.\n",
2044                                  map->name, sz);
2045                         if (map->def.key_size && map->def.key_size != sz) {
2046                                 pr_warn("map '%s': conflicting key size %u != %u.\n",
2047                                         map->name, map->def.key_size, sz);
2048                                 return -EINVAL;
2049                         }
2050                         map->def.key_size = sz;
2051                 } else if (strcmp(name, "key") == 0) {
2052                         __s64 sz;
2053
2054                         t = btf__type_by_id(obj->btf, m->type);
2055                         if (!t) {
2056                                 pr_warn("map '%s': key type [%d] not found.\n",
2057                                         map->name, m->type);
2058                                 return -EINVAL;
2059                         }
2060                         if (!btf_is_ptr(t)) {
2061                                 pr_warn("map '%s': key spec is not PTR: %s.\n",
2062                                         map->name, btf_kind_str(t));
2063                                 return -EINVAL;
2064                         }
2065                         sz = btf__resolve_size(obj->btf, t->type);
2066                         if (sz < 0) {
2067                                 pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
2068                                         map->name, t->type, (ssize_t)sz);
2069                                 return sz;
2070                         }
2071                         pr_debug("map '%s': found key [%u], sz = %zd.\n",
2072                                  map->name, t->type, (ssize_t)sz);
2073                         if (map->def.key_size && map->def.key_size != sz) {
2074                                 pr_warn("map '%s': conflicting key size %u != %zd.\n",
2075                                         map->name, map->def.key_size, (ssize_t)sz);
2076                                 return -EINVAL;
2077                         }
2078                         map->def.key_size = sz;
2079                         map->btf_key_type_id = t->type;
2080                 } else if (strcmp(name, "value_size") == 0) {
2081                         __u32 sz;
2082
2083                         if (!get_map_field_int(map->name, obj->btf, m, &sz))
2084                                 return -EINVAL;
2085                         pr_debug("map '%s': found value_size = %u.\n",
2086                                  map->name, sz);
2087                         if (map->def.value_size && map->def.value_size != sz) {
2088                                 pr_warn("map '%s': conflicting value size %u != %u.\n",
2089                                         map->name, map->def.value_size, sz);
2090                                 return -EINVAL;
2091                         }
2092                         map->def.value_size = sz;
2093                 } else if (strcmp(name, "value") == 0) {
2094                         __s64 sz;
2095
2096                         t = btf__type_by_id(obj->btf, m->type);
2097                         if (!t) {
2098                                 pr_warn("map '%s': value type [%d] not found.\n",
2099                                         map->name, m->type);
2100                                 return -EINVAL;
2101                         }
2102                         if (!btf_is_ptr(t)) {
2103                                 pr_warn("map '%s': value spec is not PTR: %s.\n",
2104                                         map->name, btf_kind_str(t));
2105                                 return -EINVAL;
2106                         }
2107                         sz = btf__resolve_size(obj->btf, t->type);
2108                         if (sz < 0) {
2109                                 pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
2110                                         map->name, t->type, (ssize_t)sz);
2111                                 return sz;
2112                         }
2113                         pr_debug("map '%s': found value [%u], sz = %zd.\n",
2114                                  map->name, t->type, (ssize_t)sz);
2115                         if (map->def.value_size && map->def.value_size != sz) {
2116                                 pr_warn("map '%s': conflicting value size %u != %zd.\n",
2117                                         map->name, map->def.value_size, (ssize_t)sz);
2118                                 return -EINVAL;
2119                         }
2120                         map->def.value_size = sz;
2121                         map->btf_value_type_id = t->type;
2122                 }
2123                 else if (strcmp(name, "values") == 0) {
2124                         int err;
2125
2126                         if (is_inner) {
2127                                 pr_warn("map '%s': multi-level inner maps not supported.\n",
2128                                         map->name);
2129                                 return -ENOTSUP;
2130                         }
2131                         if (i != vlen - 1) {
2132                                 pr_warn("map '%s': '%s' member should be last.\n",
2133                                         map->name, name);
2134                                 return -EINVAL;
2135                         }
2136                         if (!bpf_map_type__is_map_in_map(map->def.type)) {
2137                                 pr_warn("map '%s': should be map-in-map.\n",
2138                                         map->name);
2139                                 return -ENOTSUP;
2140                         }
2141                         if (map->def.value_size && map->def.value_size != 4) {
2142                                 pr_warn("map '%s': conflicting value size %u != 4.\n",
2143                                         map->name, map->def.value_size);
2144                                 return -EINVAL;
2145                         }
2146                         map->def.value_size = 4;
2147                         t = btf__type_by_id(obj->btf, m->type);
2148                         if (!t) {
2149                                 pr_warn("map '%s': map-in-map inner type [%d] not found.\n",
2150                                         map->name, m->type);
2151                                 return -EINVAL;
2152                         }
2153                         if (!btf_is_array(t) || btf_array(t)->nelems) {
2154                                 pr_warn("map '%s': map-in-map inner spec is not a zero-sized array.\n",
2155                                         map->name);
2156                                 return -EINVAL;
2157                         }
2158                         t = skip_mods_and_typedefs(obj->btf, btf_array(t)->type,
2159                                                    NULL);
2160                         if (!btf_is_ptr(t)) {
2161                                 pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n",
2162                                         map->name, btf_kind_str(t));
2163                                 return -EINVAL;
2164                         }
2165                         t = skip_mods_and_typedefs(obj->btf, t->type, NULL);
2166                         if (!btf_is_struct(t)) {
2167                                 pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n",
2168                                         map->name, btf_kind_str(t));
2169                                 return -EINVAL;
2170                         }
2171
2172                         map->inner_map = calloc(1, sizeof(*map->inner_map));
2173                         if (!map->inner_map)
2174                                 return -ENOMEM;
2175                         map->inner_map->sec_idx = obj->efile.btf_maps_shndx;
2176                         map->inner_map->name = malloc(strlen(map->name) +
2177                                                       sizeof(".inner") + 1);
2178                         if (!map->inner_map->name)
2179                                 return -ENOMEM;
2180                         sprintf(map->inner_map->name, "%s.inner", map->name);
2181
2182                         err = parse_btf_map_def(obj, map->inner_map, t, strict,
2183                                                 true /* is_inner */, NULL);
2184                         if (err)
2185                                 return err;
2186                 } else if (strcmp(name, "pinning") == 0) {
2187                         __u32 val;
2188                         int err;
2189
2190                         if (is_inner) {
2191                                 pr_debug("map '%s': inner def can't be pinned.\n",
2192                                          map->name);
2193                                 return -EINVAL;
2194                         }
2195                         if (!get_map_field_int(map->name, obj->btf, m, &val))
2196                                 return -EINVAL;
2197                         pr_debug("map '%s': found pinning = %u.\n",
2198                                  map->name, val);
2199
2200                         if (val != LIBBPF_PIN_NONE &&
2201                             val != LIBBPF_PIN_BY_NAME) {
2202                                 pr_warn("map '%s': invalid pinning value %u.\n",
2203                                         map->name, val);
2204                                 return -EINVAL;
2205                         }
2206                         if (val == LIBBPF_PIN_BY_NAME) {
2207                                 err = build_map_pin_path(map, pin_root_path);
2208                                 if (err) {
2209                                         pr_warn("map '%s': couldn't build pin path.\n",
2210                                                 map->name);
2211                                         return err;
2212                                 }
2213                         }
2214                 } else {
2215                         if (strict) {
2216                                 pr_warn("map '%s': unknown field '%s'.\n",
2217                                         map->name, name);
2218                                 return -ENOTSUP;
2219                         }
2220                         pr_debug("map '%s': ignoring unknown field '%s'.\n",
2221                                  map->name, name);
2222                 }
2223         }
2224
2225         if (map->def.type == BPF_MAP_TYPE_UNSPEC) {
2226                 pr_warn("map '%s': map type isn't specified.\n", map->name);
2227                 return -EINVAL;
2228         }
2229
2230         return 0;
2231 }
2232
2233 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
2234                                          const struct btf_type *sec,
2235                                          int var_idx, int sec_idx,
2236                                          const Elf_Data *data, bool strict,
2237                                          const char *pin_root_path)
2238 {
2239         const struct btf_type *var, *def;
2240         const struct btf_var_secinfo *vi;
2241         const struct btf_var *var_extra;
2242         const char *map_name;
2243         struct bpf_map *map;
2244
2245         vi = btf_var_secinfos(sec) + var_idx;
2246         var = btf__type_by_id(obj->btf, vi->type);
2247         var_extra = btf_var(var);
2248         map_name = btf__name_by_offset(obj->btf, var->name_off);
2249
2250         if (map_name == NULL || map_name[0] == '\0') {
2251                 pr_warn("map #%d: empty name.\n", var_idx);
2252                 return -EINVAL;
2253         }
2254         if ((__u64)vi->offset + vi->size > data->d_size) {
2255                 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
2256                 return -EINVAL;
2257         }
2258         if (!btf_is_var(var)) {
2259                 pr_warn("map '%s': unexpected var kind %s.\n",
2260                         map_name, btf_kind_str(var));
2261                 return -EINVAL;
2262         }
2263         if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2264             var_extra->linkage != BTF_VAR_STATIC) {
2265                 pr_warn("map '%s': unsupported var linkage %u.\n",
2266                         map_name, var_extra->linkage);
2267                 return -EOPNOTSUPP;
2268         }
2269
2270         def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
2271         if (!btf_is_struct(def)) {
2272                 pr_warn("map '%s': unexpected def kind %s.\n",
2273                         map_name, btf_kind_str(var));
2274                 return -EINVAL;
2275         }
2276         if (def->size > vi->size) {
2277                 pr_warn("map '%s': invalid def size.\n", map_name);
2278                 return -EINVAL;
2279         }
2280
2281         map = bpf_object__add_map(obj);
2282         if (IS_ERR(map))
2283                 return PTR_ERR(map);
2284         map->name = strdup(map_name);
2285         if (!map->name) {
2286                 pr_warn("map '%s': failed to alloc map name.\n", map_name);
2287                 return -ENOMEM;
2288         }
2289         map->libbpf_type = LIBBPF_MAP_UNSPEC;
2290         map->def.type = BPF_MAP_TYPE_UNSPEC;
2291         map->sec_idx = sec_idx;
2292         map->sec_offset = vi->offset;
2293         map->btf_var_idx = var_idx;
2294         pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
2295                  map_name, map->sec_idx, map->sec_offset);
2296
2297         return parse_btf_map_def(obj, map, def, strict, false, pin_root_path);
2298 }
2299
2300 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
2301                                           const char *pin_root_path)
2302 {
2303         const struct btf_type *sec = NULL;
2304         int nr_types, i, vlen, err;
2305         const struct btf_type *t;
2306         const char *name;
2307         Elf_Data *data;
2308         Elf_Scn *scn;
2309
2310         if (obj->efile.btf_maps_shndx < 0)
2311                 return 0;
2312
2313         scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx);
2314         data = elf_sec_data(obj, scn);
2315         if (!scn || !data) {
2316                 pr_warn("elf: failed to get %s map definitions for %s\n",
2317                         MAPS_ELF_SEC, obj->path);
2318                 return -EINVAL;
2319         }
2320
2321         nr_types = btf__get_nr_types(obj->btf);
2322         for (i = 1; i <= nr_types; i++) {
2323                 t = btf__type_by_id(obj->btf, i);
2324                 if (!btf_is_datasec(t))
2325                         continue;
2326                 name = btf__name_by_offset(obj->btf, t->name_off);
2327                 if (strcmp(name, MAPS_ELF_SEC) == 0) {
2328                         sec = t;
2329                         obj->efile.btf_maps_sec_btf_id = i;
2330                         break;
2331                 }
2332         }
2333
2334         if (!sec) {
2335                 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
2336                 return -ENOENT;
2337         }
2338
2339         vlen = btf_vlen(sec);
2340         for (i = 0; i < vlen; i++) {
2341                 err = bpf_object__init_user_btf_map(obj, sec, i,
2342                                                     obj->efile.btf_maps_shndx,
2343                                                     data, strict,
2344                                                     pin_root_path);
2345                 if (err)
2346                         return err;
2347         }
2348
2349         return 0;
2350 }
2351
2352 static int bpf_object__init_maps(struct bpf_object *obj,
2353                                  const struct bpf_object_open_opts *opts)
2354 {
2355         const char *pin_root_path;
2356         bool strict;
2357         int err;
2358
2359         strict = !OPTS_GET(opts, relaxed_maps, false);
2360         pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
2361
2362         err = bpf_object__init_user_maps(obj, strict);
2363         err = err ?: bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
2364         err = err ?: bpf_object__init_global_data_maps(obj);
2365         err = err ?: bpf_object__init_kconfig_map(obj);
2366         err = err ?: bpf_object__init_struct_ops_maps(obj);
2367         if (err)
2368                 return err;
2369
2370         return 0;
2371 }
2372
2373 static bool section_have_execinstr(struct bpf_object *obj, int idx)
2374 {
2375         GElf_Shdr sh;
2376
2377         if (elf_sec_hdr(obj, elf_sec_by_idx(obj, idx), &sh))
2378                 return false;
2379
2380         return sh.sh_flags & SHF_EXECINSTR;
2381 }
2382
2383 static bool btf_needs_sanitization(struct bpf_object *obj)
2384 {
2385         bool has_func_global = kernel_supports(FEAT_BTF_GLOBAL_FUNC);
2386         bool has_datasec = kernel_supports(FEAT_BTF_DATASEC);
2387         bool has_func = kernel_supports(FEAT_BTF_FUNC);
2388
2389         return !has_func || !has_datasec || !has_func_global;
2390 }
2391
2392 static void bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf)
2393 {
2394         bool has_func_global = kernel_supports(FEAT_BTF_GLOBAL_FUNC);
2395         bool has_datasec = kernel_supports(FEAT_BTF_DATASEC);
2396         bool has_func = kernel_supports(FEAT_BTF_FUNC);
2397         struct btf_type *t;
2398         int i, j, vlen;
2399
2400         for (i = 1; i <= btf__get_nr_types(btf); i++) {
2401                 t = (struct btf_type *)btf__type_by_id(btf, i);
2402
2403                 if (!has_datasec && btf_is_var(t)) {
2404                         /* replace VAR with INT */
2405                         t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
2406                         /*
2407                          * using size = 1 is the safest choice, 4 will be too
2408                          * big and cause kernel BTF validation failure if
2409                          * original variable took less than 4 bytes
2410                          */
2411                         t->size = 1;
2412                         *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
2413                 } else if (!has_datasec && btf_is_datasec(t)) {
2414                         /* replace DATASEC with STRUCT */
2415                         const struct btf_var_secinfo *v = btf_var_secinfos(t);
2416                         struct btf_member *m = btf_members(t);
2417                         struct btf_type *vt;
2418                         char *name;
2419
2420                         name = (char *)btf__name_by_offset(btf, t->name_off);
2421                         while (*name) {
2422                                 if (*name == '.')
2423                                         *name = '_';
2424                                 name++;
2425                         }
2426
2427                         vlen = btf_vlen(t);
2428                         t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
2429                         for (j = 0; j < vlen; j++, v++, m++) {
2430                                 /* order of field assignments is important */
2431                                 m->offset = v->offset * 8;
2432                                 m->type = v->type;
2433                                 /* preserve variable name as member name */
2434                                 vt = (void *)btf__type_by_id(btf, v->type);
2435                                 m->name_off = vt->name_off;
2436                         }
2437                 } else if (!has_func && btf_is_func_proto(t)) {
2438                         /* replace FUNC_PROTO with ENUM */
2439                         vlen = btf_vlen(t);
2440                         t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
2441                         t->size = sizeof(__u32); /* kernel enforced */
2442                 } else if (!has_func && btf_is_func(t)) {
2443                         /* replace FUNC with TYPEDEF */
2444                         t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
2445                 } else if (!has_func_global && btf_is_func(t)) {
2446                         /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */
2447                         t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0);
2448                 }
2449         }
2450 }
2451
2452 static bool libbpf_needs_btf(const struct bpf_object *obj)
2453 {
2454         return obj->efile.btf_maps_shndx >= 0 ||
2455                obj->efile.st_ops_shndx >= 0 ||
2456                obj->nr_extern > 0;
2457 }
2458
2459 static bool kernel_needs_btf(const struct bpf_object *obj)
2460 {
2461         return obj->efile.st_ops_shndx >= 0;
2462 }
2463
2464 static int bpf_object__init_btf(struct bpf_object *obj,
2465                                 Elf_Data *btf_data,
2466                                 Elf_Data *btf_ext_data)
2467 {
2468         int err = -ENOENT;
2469
2470         if (btf_data) {
2471                 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
2472                 if (IS_ERR(obj->btf)) {
2473                         err = PTR_ERR(obj->btf);
2474                         obj->btf = NULL;
2475                         pr_warn("Error loading ELF section %s: %d.\n",
2476                                 BTF_ELF_SEC, err);
2477                         goto out;
2478                 }
2479                 /* enforce 8-byte pointers for BPF-targeted BTFs */
2480                 btf__set_pointer_size(obj->btf, 8);
2481                 err = 0;
2482         }
2483         if (btf_ext_data) {
2484                 if (!obj->btf) {
2485                         pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
2486                                  BTF_EXT_ELF_SEC, BTF_ELF_SEC);
2487                         goto out;
2488                 }
2489                 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf,
2490                                             btf_ext_data->d_size);
2491                 if (IS_ERR(obj->btf_ext)) {
2492                         pr_warn("Error loading ELF section %s: %ld. Ignored and continue.\n",
2493                                 BTF_EXT_ELF_SEC, PTR_ERR(obj->btf_ext));
2494                         obj->btf_ext = NULL;
2495                         goto out;
2496                 }
2497         }
2498 out:
2499         if (err && libbpf_needs_btf(obj)) {
2500                 pr_warn("BTF is required, but is missing or corrupted.\n");
2501                 return err;
2502         }
2503         return 0;
2504 }
2505
2506 static int bpf_object__finalize_btf(struct bpf_object *obj)
2507 {
2508         int err;
2509
2510         if (!obj->btf)
2511                 return 0;
2512
2513         err = btf__finalize_data(obj, obj->btf);
2514         if (err) {
2515                 pr_warn("Error finalizing %s: %d.\n", BTF_ELF_SEC, err);
2516                 return err;
2517         }
2518
2519         return 0;
2520 }
2521
2522 static bool prog_needs_vmlinux_btf(struct bpf_program *prog)
2523 {
2524         if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
2525             prog->type == BPF_PROG_TYPE_LSM)
2526                 return true;
2527
2528         /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
2529          * also need vmlinux BTF
2530          */
2531         if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
2532                 return true;
2533
2534         return false;
2535 }
2536
2537 static bool obj_needs_vmlinux_btf(const struct bpf_object *obj)
2538 {
2539         struct bpf_program *prog;
2540         int i;
2541
2542         /* CO-RE relocations need kernel BTF */
2543         if (obj->btf_ext && obj->btf_ext->core_relo_info.len)
2544                 return true;
2545
2546         /* Support for typed ksyms needs kernel BTF */
2547         for (i = 0; i < obj->nr_extern; i++) {
2548                 const struct extern_desc *ext;
2549
2550                 ext = &obj->externs[i];
2551                 if (ext->type == EXT_KSYM && ext->ksym.type_id)
2552                         return true;
2553         }
2554
2555         bpf_object__for_each_program(prog, obj) {
2556                 if (!prog->load)
2557                         continue;
2558                 if (prog_needs_vmlinux_btf(prog))
2559                         return true;
2560         }
2561
2562         return false;
2563 }
2564
2565 static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force)
2566 {
2567         int err;
2568
2569         /* btf_vmlinux could be loaded earlier */
2570         if (obj->btf_vmlinux)
2571                 return 0;
2572
2573         if (!force && !obj_needs_vmlinux_btf(obj))
2574                 return 0;
2575
2576         obj->btf_vmlinux = libbpf_find_kernel_btf();
2577         if (IS_ERR(obj->btf_vmlinux)) {
2578                 err = PTR_ERR(obj->btf_vmlinux);
2579                 pr_warn("Error loading vmlinux BTF: %d\n", err);
2580                 obj->btf_vmlinux = NULL;
2581                 return err;
2582         }
2583         return 0;
2584 }
2585
2586 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
2587 {
2588         struct btf *kern_btf = obj->btf;
2589         bool btf_mandatory, sanitize;
2590         int err = 0;
2591
2592         if (!obj->btf)
2593                 return 0;
2594
2595         if (!kernel_supports(FEAT_BTF)) {
2596                 if (kernel_needs_btf(obj)) {
2597                         err = -EOPNOTSUPP;
2598                         goto report;
2599                 }
2600                 pr_debug("Kernel doesn't support BTF, skipping uploading it.\n");
2601                 return 0;
2602         }
2603
2604         sanitize = btf_needs_sanitization(obj);
2605         if (sanitize) {
2606                 const void *raw_data;
2607                 __u32 sz;
2608
2609                 /* clone BTF to sanitize a copy and leave the original intact */
2610                 raw_data = btf__get_raw_data(obj->btf, &sz);
2611                 kern_btf = btf__new(raw_data, sz);
2612                 if (IS_ERR(kern_btf))
2613                         return PTR_ERR(kern_btf);
2614
2615                 /* enforce 8-byte pointers for BPF-targeted BTFs */
2616                 btf__set_pointer_size(obj->btf, 8);
2617                 bpf_object__sanitize_btf(obj, kern_btf);
2618         }
2619
2620         err = btf__load(kern_btf);
2621         if (sanitize) {
2622                 if (!err) {
2623                         /* move fd to libbpf's BTF */
2624                         btf__set_fd(obj->btf, btf__fd(kern_btf));
2625                         btf__set_fd(kern_btf, -1);
2626                 }
2627                 btf__free(kern_btf);
2628         }
2629 report:
2630         if (err) {
2631                 btf_mandatory = kernel_needs_btf(obj);
2632                 pr_warn("Error loading .BTF into kernel: %d. %s\n", err,
2633                         btf_mandatory ? "BTF is mandatory, can't proceed."
2634                                       : "BTF is optional, ignoring.");
2635                 if (!btf_mandatory)
2636                         err = 0;
2637         }
2638         return err;
2639 }
2640
2641 static const char *elf_sym_str(const struct bpf_object *obj, size_t off)
2642 {
2643         const char *name;
2644
2645         name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off);
2646         if (!name) {
2647                 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
2648                         off, obj->path, elf_errmsg(-1));
2649                 return NULL;
2650         }
2651
2652         return name;
2653 }
2654
2655 static const char *elf_sec_str(const struct bpf_object *obj, size_t off)
2656 {
2657         const char *name;
2658
2659         name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off);
2660         if (!name) {
2661                 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
2662                         off, obj->path, elf_errmsg(-1));
2663                 return NULL;
2664         }
2665
2666         return name;
2667 }
2668
2669 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx)
2670 {
2671         Elf_Scn *scn;
2672
2673         scn = elf_getscn(obj->efile.elf, idx);
2674         if (!scn) {
2675                 pr_warn("elf: failed to get section(%zu) from %s: %s\n",
2676                         idx, obj->path, elf_errmsg(-1));
2677                 return NULL;
2678         }
2679         return scn;
2680 }
2681
2682 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name)
2683 {
2684         Elf_Scn *scn = NULL;
2685         Elf *elf = obj->efile.elf;
2686         const char *sec_name;
2687
2688         while ((scn = elf_nextscn(elf, scn)) != NULL) {
2689                 sec_name = elf_sec_name(obj, scn);
2690                 if (!sec_name)
2691                         return NULL;
2692
2693                 if (strcmp(sec_name, name) != 0)
2694                         continue;
2695
2696                 return scn;
2697         }
2698         return NULL;
2699 }
2700
2701 static int elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn, GElf_Shdr *hdr)
2702 {
2703         if (!scn)
2704                 return -EINVAL;
2705
2706         if (gelf_getshdr(scn, hdr) != hdr) {
2707                 pr_warn("elf: failed to get section(%zu) header from %s: %s\n",
2708                         elf_ndxscn(scn), obj->path, elf_errmsg(-1));
2709                 return -EINVAL;
2710         }
2711
2712         return 0;
2713 }
2714
2715 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn)
2716 {
2717         const char *name;
2718         GElf_Shdr sh;
2719
2720         if (!scn)
2721                 return NULL;
2722
2723         if (elf_sec_hdr(obj, scn, &sh))
2724                 return NULL;
2725
2726         name = elf_sec_str(obj, sh.sh_name);
2727         if (!name) {
2728                 pr_warn("elf: failed to get section(%zu) name from %s: %s\n",
2729                         elf_ndxscn(scn), obj->path, elf_errmsg(-1));
2730                 return NULL;
2731         }
2732
2733         return name;
2734 }
2735
2736 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn)
2737 {
2738         Elf_Data *data;
2739
2740         if (!scn)
2741                 return NULL;
2742
2743         data = elf_getdata(scn, 0);
2744         if (!data) {
2745                 pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n",
2746                         elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>",
2747                         obj->path, elf_errmsg(-1));
2748                 return NULL;
2749         }
2750
2751         return data;
2752 }
2753
2754 static int elf_sym_by_sec_off(const struct bpf_object *obj, size_t sec_idx,
2755                               size_t off, __u32 sym_type, GElf_Sym *sym)
2756 {
2757         Elf_Data *symbols = obj->efile.symbols;
2758         size_t n = symbols->d_size / sizeof(GElf_Sym);
2759         int i;
2760
2761         for (i = 0; i < n; i++) {
2762                 if (!gelf_getsym(symbols, i, sym))
2763                         continue;
2764                 if (sym->st_shndx != sec_idx || sym->st_value != off)
2765                         continue;
2766                 if (GELF_ST_TYPE(sym->st_info) != sym_type)
2767                         continue;
2768                 return 0;
2769         }
2770
2771         return -ENOENT;
2772 }
2773
2774 static bool is_sec_name_dwarf(const char *name)
2775 {
2776         /* approximation, but the actual list is too long */
2777         return strncmp(name, ".debug_", sizeof(".debug_") - 1) == 0;
2778 }
2779
2780 static bool ignore_elf_section(GElf_Shdr *hdr, const char *name)
2781 {
2782         /* no special handling of .strtab */
2783         if (hdr->sh_type == SHT_STRTAB)
2784                 return true;
2785
2786         /* ignore .llvm_addrsig section as well */
2787         if (hdr->sh_type == 0x6FFF4C03 /* SHT_LLVM_ADDRSIG */)
2788                 return true;
2789
2790         /* no subprograms will lead to an empty .text section, ignore it */
2791         if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 &&
2792             strcmp(name, ".text") == 0)
2793                 return true;
2794
2795         /* DWARF sections */
2796         if (is_sec_name_dwarf(name))
2797                 return true;
2798
2799         if (strncmp(name, ".rel", sizeof(".rel") - 1) == 0) {
2800                 name += sizeof(".rel") - 1;
2801                 /* DWARF section relocations */
2802                 if (is_sec_name_dwarf(name))
2803                         return true;
2804
2805                 /* .BTF and .BTF.ext don't need relocations */
2806                 if (strcmp(name, BTF_ELF_SEC) == 0 ||
2807                     strcmp(name, BTF_EXT_ELF_SEC) == 0)
2808                         return true;
2809         }
2810
2811         return false;
2812 }
2813
2814 static int cmp_progs(const void *_a, const void *_b)
2815 {
2816         const struct bpf_program *a = _a;
2817         const struct bpf_program *b = _b;
2818
2819         if (a->sec_idx != b->sec_idx)
2820                 return a->sec_idx < b->sec_idx ? -1 : 1;
2821
2822         /* sec_insn_off can't be the same within the section */
2823         return a->sec_insn_off < b->sec_insn_off ? -1 : 1;
2824 }
2825
2826 static int bpf_object__elf_collect(struct bpf_object *obj)
2827 {
2828         Elf *elf = obj->efile.elf;
2829         Elf_Data *btf_ext_data = NULL;
2830         Elf_Data *btf_data = NULL;
2831         int idx = 0, err = 0;
2832         const char *name;
2833         Elf_Data *data;
2834         Elf_Scn *scn;
2835         GElf_Shdr sh;
2836
2837         /* a bunch of ELF parsing functionality depends on processing symbols,
2838          * so do the first pass and find the symbol table
2839          */
2840         scn = NULL;
2841         while ((scn = elf_nextscn(elf, scn)) != NULL) {
2842                 if (elf_sec_hdr(obj, scn, &sh))
2843                         return -LIBBPF_ERRNO__FORMAT;
2844
2845                 if (sh.sh_type == SHT_SYMTAB) {
2846                         if (obj->efile.symbols) {
2847                                 pr_warn("elf: multiple symbol tables in %s\n", obj->path);
2848                                 return -LIBBPF_ERRNO__FORMAT;
2849                         }
2850
2851                         data = elf_sec_data(obj, scn);
2852                         if (!data)
2853                                 return -LIBBPF_ERRNO__FORMAT;
2854
2855                         obj->efile.symbols = data;
2856                         obj->efile.symbols_shndx = elf_ndxscn(scn);
2857                         obj->efile.strtabidx = sh.sh_link;
2858                 }
2859         }
2860
2861         scn = NULL;
2862         while ((scn = elf_nextscn(elf, scn)) != NULL) {
2863                 idx++;
2864
2865                 if (elf_sec_hdr(obj, scn, &sh))
2866                         return -LIBBPF_ERRNO__FORMAT;
2867
2868                 name = elf_sec_str(obj, sh.sh_name);
2869                 if (!name)
2870                         return -LIBBPF_ERRNO__FORMAT;
2871
2872                 if (ignore_elf_section(&sh, name))
2873                         continue;
2874
2875                 data = elf_sec_data(obj, scn);
2876                 if (!data)
2877                         return -LIBBPF_ERRNO__FORMAT;
2878
2879                 pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
2880                          idx, name, (unsigned long)data->d_size,
2881                          (int)sh.sh_link, (unsigned long)sh.sh_flags,
2882                          (int)sh.sh_type);
2883
2884                 if (strcmp(name, "license") == 0) {
2885                         err = bpf_object__init_license(obj, data->d_buf, data->d_size);
2886                         if (err)
2887                                 return err;
2888                 } else if (strcmp(name, "version") == 0) {
2889                         err = bpf_object__init_kversion(obj, data->d_buf, data->d_size);
2890                         if (err)
2891                                 return err;
2892                 } else if (strcmp(name, "maps") == 0) {
2893                         obj->efile.maps_shndx = idx;
2894                 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
2895                         obj->efile.btf_maps_shndx = idx;
2896                 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
2897                         btf_data = data;
2898                 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
2899                         btf_ext_data = data;
2900                 } else if (sh.sh_type == SHT_SYMTAB) {
2901                         /* already processed during the first pass above */
2902                 } else if (sh.sh_type == SHT_PROGBITS && data->d_size > 0) {
2903                         if (sh.sh_flags & SHF_EXECINSTR) {
2904                                 if (strcmp(name, ".text") == 0)
2905                                         obj->efile.text_shndx = idx;
2906                                 err = bpf_object__add_programs(obj, data, name, idx);
2907                                 if (err)
2908                                         return err;
2909                         } else if (strcmp(name, DATA_SEC) == 0) {
2910                                 obj->efile.data = data;
2911                                 obj->efile.data_shndx = idx;
2912                         } else if (strcmp(name, RODATA_SEC) == 0) {
2913                                 obj->efile.rodata = data;
2914                                 obj->efile.rodata_shndx = idx;
2915                         } else if (strcmp(name, STRUCT_OPS_SEC) == 0) {
2916                                 obj->efile.st_ops_data = data;
2917                                 obj->efile.st_ops_shndx = idx;
2918                         } else {
2919                                 pr_info("elf: skipping unrecognized data section(%d) %s\n",
2920                                         idx, name);
2921                         }
2922                 } else if (sh.sh_type == SHT_REL) {
2923                         int nr_sects = obj->efile.nr_reloc_sects;
2924                         void *sects = obj->efile.reloc_sects;
2925                         int sec = sh.sh_info; /* points to other section */
2926
2927                         /* Only do relo for section with exec instructions */
2928                         if (!section_have_execinstr(obj, sec) &&
2929                             strcmp(name, ".rel" STRUCT_OPS_SEC) &&
2930                             strcmp(name, ".rel" MAPS_ELF_SEC)) {
2931                                 pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n",
2932                                         idx, name, sec,
2933                                         elf_sec_name(obj, elf_sec_by_idx(obj, sec)) ?: "<?>");
2934                                 continue;
2935                         }
2936
2937                         sects = libbpf_reallocarray(sects, nr_sects + 1,
2938                                                     sizeof(*obj->efile.reloc_sects));
2939                         if (!sects)
2940                                 return -ENOMEM;
2941
2942                         obj->efile.reloc_sects = sects;
2943                         obj->efile.nr_reloc_sects++;
2944
2945                         obj->efile.reloc_sects[nr_sects].shdr = sh;
2946                         obj->efile.reloc_sects[nr_sects].data = data;
2947                 } else if (sh.sh_type == SHT_NOBITS && strcmp(name, BSS_SEC) == 0) {
2948                         obj->efile.bss = data;
2949                         obj->efile.bss_shndx = idx;
2950                 } else {
2951                         pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name,
2952                                 (size_t)sh.sh_size);
2953                 }
2954         }
2955
2956         if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
2957                 pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path);
2958                 return -LIBBPF_ERRNO__FORMAT;
2959         }
2960
2961         /* sort BPF programs by section name and in-section instruction offset
2962          * for faster search */
2963         qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs);
2964
2965         return bpf_object__init_btf(obj, btf_data, btf_ext_data);
2966 }
2967
2968 static bool sym_is_extern(const GElf_Sym *sym)
2969 {
2970         int bind = GELF_ST_BIND(sym->st_info);
2971         /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
2972         return sym->st_shndx == SHN_UNDEF &&
2973                (bind == STB_GLOBAL || bind == STB_WEAK) &&
2974                GELF_ST_TYPE(sym->st_info) == STT_NOTYPE;
2975 }
2976
2977 static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
2978 {
2979         const struct btf_type *t;
2980         const char *var_name;
2981         int i, n;
2982
2983         if (!btf)
2984                 return -ESRCH;
2985
2986         n = btf__get_nr_types(btf);
2987         for (i = 1; i <= n; i++) {
2988                 t = btf__type_by_id(btf, i);
2989
2990                 if (!btf_is_var(t))
2991                         continue;
2992
2993                 var_name = btf__name_by_offset(btf, t->name_off);
2994                 if (strcmp(var_name, ext_name))
2995                         continue;
2996
2997                 if (btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
2998                         return -EINVAL;
2999
3000                 return i;
3001         }
3002
3003         return -ENOENT;
3004 }
3005
3006 static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) {
3007         const struct btf_var_secinfo *vs;
3008         const struct btf_type *t;
3009         int i, j, n;
3010
3011         if (!btf)
3012                 return -ESRCH;
3013
3014         n = btf__get_nr_types(btf);
3015         for (i = 1; i <= n; i++) {
3016                 t = btf__type_by_id(btf, i);
3017
3018                 if (!btf_is_datasec(t))
3019                         continue;
3020
3021                 vs = btf_var_secinfos(t);
3022                 for (j = 0; j < btf_vlen(t); j++, vs++) {
3023                         if (vs->type == ext_btf_id)
3024                                 return i;
3025                 }
3026         }
3027
3028         return -ENOENT;
3029 }
3030
3031 static enum kcfg_type find_kcfg_type(const struct btf *btf, int id,
3032                                      bool *is_signed)
3033 {
3034         const struct btf_type *t;
3035         const char *name;
3036
3037         t = skip_mods_and_typedefs(btf, id, NULL);
3038         name = btf__name_by_offset(btf, t->name_off);
3039
3040         if (is_signed)
3041                 *is_signed = false;
3042         switch (btf_kind(t)) {
3043         case BTF_KIND_INT: {
3044                 int enc = btf_int_encoding(t);
3045
3046                 if (enc & BTF_INT_BOOL)
3047                         return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN;
3048                 if (is_signed)
3049                         *is_signed = enc & BTF_INT_SIGNED;
3050                 if (t->size == 1)
3051                         return KCFG_CHAR;
3052                 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
3053                         return KCFG_UNKNOWN;
3054                 return KCFG_INT;
3055         }
3056         case BTF_KIND_ENUM:
3057                 if (t->size != 4)
3058                         return KCFG_UNKNOWN;
3059                 if (strcmp(name, "libbpf_tristate"))
3060                         return KCFG_UNKNOWN;
3061                 return KCFG_TRISTATE;
3062         case BTF_KIND_ARRAY:
3063                 if (btf_array(t)->nelems == 0)
3064                         return KCFG_UNKNOWN;
3065                 if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR)
3066                         return KCFG_UNKNOWN;
3067                 return KCFG_CHAR_ARR;
3068         default:
3069                 return KCFG_UNKNOWN;
3070         }
3071 }
3072
3073 static int cmp_externs(const void *_a, const void *_b)
3074 {
3075         const struct extern_desc *a = _a;
3076         const struct extern_desc *b = _b;
3077
3078         if (a->type != b->type)
3079                 return a->type < b->type ? -1 : 1;
3080
3081         if (a->type == EXT_KCFG) {
3082                 /* descending order by alignment requirements */
3083                 if (a->kcfg.align != b->kcfg.align)
3084                         return a->kcfg.align > b->kcfg.align ? -1 : 1;
3085                 /* ascending order by size, within same alignment class */
3086                 if (a->kcfg.sz != b->kcfg.sz)
3087                         return a->kcfg.sz < b->kcfg.sz ? -1 : 1;
3088         }
3089
3090         /* resolve ties by name */
3091         return strcmp(a->name, b->name);
3092 }
3093
3094 static int find_int_btf_id(const struct btf *btf)
3095 {
3096         const struct btf_type *t;
3097         int i, n;
3098
3099         n = btf__get_nr_types(btf);
3100         for (i = 1; i <= n; i++) {
3101                 t = btf__type_by_id(btf, i);
3102
3103                 if (btf_is_int(t) && btf_int_bits(t) == 32)
3104                         return i;
3105         }
3106
3107         return 0;
3108 }
3109
3110 static int bpf_object__collect_externs(struct bpf_object *obj)
3111 {
3112         struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
3113         const struct btf_type *t;
3114         struct extern_desc *ext;
3115         int i, n, off;
3116         const char *ext_name, *sec_name;
3117         Elf_Scn *scn;
3118         GElf_Shdr sh;
3119
3120         if (!obj->efile.symbols)
3121                 return 0;
3122
3123         scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx);
3124         if (elf_sec_hdr(obj, scn, &sh))
3125                 return -LIBBPF_ERRNO__FORMAT;
3126
3127         n = sh.sh_size / sh.sh_entsize;
3128         pr_debug("looking for externs among %d symbols...\n", n);
3129
3130         for (i = 0; i < n; i++) {
3131                 GElf_Sym sym;
3132
3133                 if (!gelf_getsym(obj->efile.symbols, i, &sym))
3134                         return -LIBBPF_ERRNO__FORMAT;
3135                 if (!sym_is_extern(&sym))
3136                         continue;
3137                 ext_name = elf_sym_str(obj, sym.st_name);
3138                 if (!ext_name || !ext_name[0])
3139                         continue;
3140
3141                 ext = obj->externs;
3142                 ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
3143                 if (!ext)
3144                         return -ENOMEM;
3145                 obj->externs = ext;
3146                 ext = &ext[obj->nr_extern];
3147                 memset(ext, 0, sizeof(*ext));
3148                 obj->nr_extern++;
3149
3150                 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
3151                 if (ext->btf_id <= 0) {
3152                         pr_warn("failed to find BTF for extern '%s': %d\n",
3153                                 ext_name, ext->btf_id);
3154                         return ext->btf_id;
3155                 }
3156                 t = btf__type_by_id(obj->btf, ext->btf_id);
3157                 ext->name = btf__name_by_offset(obj->btf, t->name_off);
3158                 ext->sym_idx = i;
3159                 ext->is_weak = GELF_ST_BIND(sym.st_info) == STB_WEAK;
3160
3161                 ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id);
3162                 if (ext->sec_btf_id <= 0) {
3163                         pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n",
3164                                 ext_name, ext->btf_id, ext->sec_btf_id);
3165                         return ext->sec_btf_id;
3166                 }
3167                 sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id);
3168                 sec_name = btf__name_by_offset(obj->btf, sec->name_off);
3169
3170                 if (strcmp(sec_name, KCONFIG_SEC) == 0) {
3171                         kcfg_sec = sec;
3172                         ext->type = EXT_KCFG;
3173                         ext->kcfg.sz = btf__resolve_size(obj->btf, t->type);
3174                         if (ext->kcfg.sz <= 0) {
3175                                 pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n",
3176                                         ext_name, ext->kcfg.sz);
3177                                 return ext->kcfg.sz;
3178                         }
3179                         ext->kcfg.align = btf__align_of(obj->btf, t->type);
3180                         if (ext->kcfg.align <= 0) {
3181                                 pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n",
3182                                         ext_name, ext->kcfg.align);
3183                                 return -EINVAL;
3184                         }
3185                         ext->kcfg.type = find_kcfg_type(obj->btf, t->type,
3186                                                         &ext->kcfg.is_signed);
3187                         if (ext->kcfg.type == KCFG_UNKNOWN) {
3188                                 pr_warn("extern (kcfg) '%s' type is unsupported\n", ext_name);
3189                                 return -ENOTSUP;
3190                         }
3191                 } else if (strcmp(sec_name, KSYMS_SEC) == 0) {
3192                         ksym_sec = sec;
3193                         ext->type = EXT_KSYM;
3194                         skip_mods_and_typedefs(obj->btf, t->type,
3195                                                &ext->ksym.type_id);
3196                 } else {
3197                         pr_warn("unrecognized extern section '%s'\n", sec_name);
3198                         return -ENOTSUP;
3199                 }
3200         }
3201         pr_debug("collected %d externs total\n", obj->nr_extern);
3202
3203         if (!obj->nr_extern)
3204                 return 0;
3205
3206         /* sort externs by type, for kcfg ones also by (align, size, name) */
3207         qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
3208
3209         /* for .ksyms section, we need to turn all externs into allocated
3210          * variables in BTF to pass kernel verification; we do this by
3211          * pretending that each extern is a 8-byte variable
3212          */
3213         if (ksym_sec) {
3214                 /* find existing 4-byte integer type in BTF to use for fake
3215                  * extern variables in DATASEC
3216                  */
3217                 int int_btf_id = find_int_btf_id(obj->btf);
3218
3219                 for (i = 0; i < obj->nr_extern; i++) {
3220                         ext = &obj->externs[i];
3221                         if (ext->type != EXT_KSYM)
3222                                 continue;
3223                         pr_debug("extern (ksym) #%d: symbol %d, name %s\n",
3224                                  i, ext->sym_idx, ext->name);
3225                 }
3226
3227                 sec = ksym_sec;
3228                 n = btf_vlen(sec);
3229                 for (i = 0, off = 0; i < n; i++, off += sizeof(int)) {
3230                         struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
3231                         struct btf_type *vt;
3232
3233                         vt = (void *)btf__type_by_id(obj->btf, vs->type);
3234                         ext_name = btf__name_by_offset(obj->btf, vt->name_off);
3235                         ext = find_extern_by_name(obj, ext_name);
3236                         if (!ext) {
3237                                 pr_warn("failed to find extern definition for BTF var '%s'\n",
3238                                         ext_name);
3239                                 return -ESRCH;
3240                         }
3241                         btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
3242                         vt->type = int_btf_id;
3243                         vs->offset = off;
3244                         vs->size = sizeof(int);
3245                 }
3246                 sec->size = off;
3247         }
3248
3249         if (kcfg_sec) {
3250                 sec = kcfg_sec;
3251                 /* for kcfg externs calculate their offsets within a .kconfig map */
3252                 off = 0;
3253                 for (i = 0; i < obj->nr_extern; i++) {
3254                         ext = &obj->externs[i];
3255                         if (ext->type != EXT_KCFG)
3256                                 continue;
3257
3258                         ext->kcfg.data_off = roundup(off, ext->kcfg.align);
3259                         off = ext->kcfg.data_off + ext->kcfg.sz;
3260                         pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n",
3261                                  i, ext->sym_idx, ext->kcfg.data_off, ext->name);
3262                 }
3263                 sec->size = off;
3264                 n = btf_vlen(sec);
3265                 for (i = 0; i < n; i++) {
3266                         struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
3267
3268                         t = btf__type_by_id(obj->btf, vs->type);
3269                         ext_name = btf__name_by_offset(obj->btf, t->name_off);
3270                         ext = find_extern_by_name(obj, ext_name);
3271                         if (!ext) {
3272                                 pr_warn("failed to find extern definition for BTF var '%s'\n",
3273                                         ext_name);
3274                                 return -ESRCH;
3275                         }
3276                         btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
3277                         vs->offset = ext->kcfg.data_off;
3278                 }
3279         }
3280         return 0;
3281 }
3282
3283 struct bpf_program *
3284 bpf_object__find_program_by_title(const struct bpf_object *obj,
3285                                   const char *title)
3286 {
3287         struct bpf_program *pos;
3288
3289         bpf_object__for_each_program(pos, obj) {
3290                 if (pos->sec_name && !strcmp(pos->sec_name, title))
3291                         return pos;
3292         }
3293         return NULL;
3294 }
3295
3296 static bool prog_is_subprog(const struct bpf_object *obj,
3297                             const struct bpf_program *prog)
3298 {
3299         /* For legacy reasons, libbpf supports an entry-point BPF programs
3300          * without SEC() attribute, i.e., those in the .text section. But if
3301          * there are 2 or more such programs in the .text section, they all
3302          * must be subprograms called from entry-point BPF programs in
3303          * designated SEC()'tions, otherwise there is no way to distinguish
3304          * which of those programs should be loaded vs which are a subprogram.
3305          * Similarly, if there is a function/program in .text and at least one
3306          * other BPF program with custom SEC() attribute, then we just assume
3307          * .text programs are subprograms (even if they are not called from
3308          * other programs), because libbpf never explicitly supported mixing
3309          * SEC()-designated BPF programs and .text entry-point BPF programs.
3310          */
3311         return prog->sec_idx == obj->efile.text_shndx && obj->nr_programs > 1;
3312 }
3313
3314 struct bpf_program *
3315 bpf_object__find_program_by_name(const struct bpf_object *obj,
3316                                  const char *name)
3317 {
3318         struct bpf_program *prog;
3319
3320         bpf_object__for_each_program(prog, obj) {
3321                 if (prog_is_subprog(obj, prog))
3322                         continue;
3323                 if (!strcmp(prog->name, name))
3324                         return prog;
3325         }
3326         return NULL;
3327 }
3328
3329 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
3330                                       int shndx)
3331 {
3332         return shndx == obj->efile.data_shndx ||
3333                shndx == obj->efile.bss_shndx ||
3334                shndx == obj->efile.rodata_shndx;
3335 }
3336
3337 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
3338                                       int shndx)
3339 {
3340         return shndx == obj->efile.maps_shndx ||
3341                shndx == obj->efile.btf_maps_shndx;
3342 }
3343
3344 static enum libbpf_map_type
3345 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
3346 {
3347         if (shndx == obj->efile.data_shndx)
3348                 return LIBBPF_MAP_DATA;
3349         else if (shndx == obj->efile.bss_shndx)
3350                 return LIBBPF_MAP_BSS;
3351         else if (shndx == obj->efile.rodata_shndx)
3352                 return LIBBPF_MAP_RODATA;
3353         else if (shndx == obj->efile.symbols_shndx)
3354                 return LIBBPF_MAP_KCONFIG;
3355         else
3356                 return LIBBPF_MAP_UNSPEC;
3357 }
3358
3359 static int bpf_program__record_reloc(struct bpf_program *prog,
3360                                      struct reloc_desc *reloc_desc,
3361                                      __u32 insn_idx, const char *sym_name,
3362                                      const GElf_Sym *sym, const GElf_Rel *rel)
3363 {
3364         struct bpf_insn *insn = &prog->insns[insn_idx];
3365         size_t map_idx, nr_maps = prog->obj->nr_maps;
3366         struct bpf_object *obj = prog->obj;
3367         __u32 shdr_idx = sym->st_shndx;
3368         enum libbpf_map_type type;
3369         const char *sym_sec_name;
3370         struct bpf_map *map;
3371
3372         reloc_desc->processed = false;
3373
3374         /* sub-program call relocation */
3375         if (insn->code == (BPF_JMP | BPF_CALL)) {
3376                 if (insn->src_reg != BPF_PSEUDO_CALL) {
3377                         pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name);
3378                         return -LIBBPF_ERRNO__RELOC;
3379                 }
3380                 /* text_shndx can be 0, if no default "main" program exists */
3381                 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
3382                         sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
3383                         pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n",
3384                                 prog->name, sym_name, sym_sec_name);
3385                         return -LIBBPF_ERRNO__RELOC;
3386                 }
3387                 if (sym->st_value % BPF_INSN_SZ) {
3388                         pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n",
3389                                 prog->name, sym_name, (size_t)sym->st_value);
3390                         return -LIBBPF_ERRNO__RELOC;
3391                 }
3392                 reloc_desc->type = RELO_CALL;
3393                 reloc_desc->insn_idx = insn_idx;
3394                 reloc_desc->sym_off = sym->st_value;
3395                 return 0;
3396         }
3397
3398         if (insn->code != (BPF_LD | BPF_IMM | BPF_DW)) {
3399                 pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n",
3400                         prog->name, sym_name, insn_idx, insn->code);
3401                 return -LIBBPF_ERRNO__RELOC;
3402         }
3403
3404         if (sym_is_extern(sym)) {
3405                 int sym_idx = GELF_R_SYM(rel->r_info);
3406                 int i, n = obj->nr_extern;
3407                 struct extern_desc *ext;
3408
3409                 for (i = 0; i < n; i++) {
3410                         ext = &obj->externs[i];
3411                         if (ext->sym_idx == sym_idx)
3412                                 break;
3413                 }
3414                 if (i >= n) {
3415                         pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n",
3416                                 prog->name, sym_name, sym_idx);
3417                         return -LIBBPF_ERRNO__RELOC;
3418                 }
3419                 pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n",
3420                          prog->name, i, ext->name, ext->sym_idx, insn_idx);
3421                 reloc_desc->type = RELO_EXTERN;
3422                 reloc_desc->insn_idx = insn_idx;
3423                 reloc_desc->sym_off = i; /* sym_off stores extern index */
3424                 return 0;
3425         }
3426
3427         if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
3428                 pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n",
3429                         prog->name, sym_name, shdr_idx);
3430                 return -LIBBPF_ERRNO__RELOC;
3431         }
3432
3433         type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
3434         sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
3435
3436         /* generic map reference relocation */
3437         if (type == LIBBPF_MAP_UNSPEC) {
3438                 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
3439                         pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n",
3440                                 prog->name, sym_name, sym_sec_name);
3441                         return -LIBBPF_ERRNO__RELOC;
3442                 }
3443                 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
3444                         map = &obj->maps[map_idx];
3445                         if (map->libbpf_type != type ||
3446                             map->sec_idx != sym->st_shndx ||
3447                             map->sec_offset != sym->st_value)
3448                                 continue;
3449                         pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n",
3450                                  prog->name, map_idx, map->name, map->sec_idx,
3451                                  map->sec_offset, insn_idx);
3452                         break;
3453                 }
3454                 if (map_idx >= nr_maps) {
3455                         pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n",
3456                                 prog->name, sym_sec_name, (size_t)sym->st_value);
3457                         return -LIBBPF_ERRNO__RELOC;
3458                 }
3459                 reloc_desc->type = RELO_LD64;
3460                 reloc_desc->insn_idx = insn_idx;
3461                 reloc_desc->map_idx = map_idx;
3462                 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
3463                 return 0;
3464         }
3465
3466         /* global data map relocation */
3467         if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
3468                 pr_warn("prog '%s': bad data relo against section '%s'\n",
3469                         prog->name, sym_sec_name);
3470                 return -LIBBPF_ERRNO__RELOC;
3471         }
3472         for (map_idx = 0; map_idx < nr_maps; map_idx++) {
3473                 map = &obj->maps[map_idx];
3474                 if (map->libbpf_type != type)
3475                         continue;
3476                 pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n",
3477                          prog->name, map_idx, map->name, map->sec_idx,
3478                          map->sec_offset, insn_idx);
3479                 break;
3480         }
3481         if (map_idx >= nr_maps) {
3482                 pr_warn("prog '%s': data relo failed to find map for section '%s'\n",
3483                         prog->name, sym_sec_name);
3484                 return -LIBBPF_ERRNO__RELOC;
3485         }
3486
3487         reloc_desc->type = RELO_DATA;
3488         reloc_desc->insn_idx = insn_idx;
3489         reloc_desc->map_idx = map_idx;
3490         reloc_desc->sym_off = sym->st_value;
3491         return 0;
3492 }
3493
3494 static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx)
3495 {
3496         return insn_idx >= prog->sec_insn_off &&
3497                insn_idx < prog->sec_insn_off + prog->sec_insn_cnt;
3498 }
3499
3500 static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj,
3501                                                  size_t sec_idx, size_t insn_idx)
3502 {
3503         int l = 0, r = obj->nr_programs - 1, m;
3504         struct bpf_program *prog;
3505
3506         while (l < r) {
3507                 m = l + (r - l + 1) / 2;
3508                 prog = &obj->programs[m];
3509
3510                 if (prog->sec_idx < sec_idx ||
3511                     (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx))
3512                         l = m;
3513                 else
3514                         r = m - 1;
3515         }
3516         /* matching program could be at index l, but it still might be the
3517          * wrong one, so we need to double check conditions for the last time
3518          */
3519         prog = &obj->programs[l];
3520         if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx))
3521                 return prog;
3522         return NULL;
3523 }
3524
3525 static int
3526 bpf_object__collect_prog_relos(struct bpf_object *obj, GElf_Shdr *shdr, Elf_Data *data)
3527 {
3528         Elf_Data *symbols = obj->efile.symbols;
3529         const char *relo_sec_name, *sec_name;
3530         size_t sec_idx = shdr->sh_info;
3531         struct bpf_program *prog;
3532         struct reloc_desc *relos;
3533         int err, i, nrels;
3534         const char *sym_name;
3535         __u32 insn_idx;
3536         GElf_Sym sym;
3537         GElf_Rel rel;
3538
3539         relo_sec_name = elf_sec_str(obj, shdr->sh_name);
3540         sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
3541         if (!relo_sec_name || !sec_name)
3542                 return -EINVAL;
3543
3544         pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n",
3545                  relo_sec_name, sec_idx, sec_name);
3546         nrels = shdr->sh_size / shdr->sh_entsize;
3547
3548         for (i = 0; i < nrels; i++) {
3549                 if (!gelf_getrel(data, i, &rel)) {
3550                         pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i);
3551                         return -LIBBPF_ERRNO__FORMAT;
3552                 }
3553                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
3554                         pr_warn("sec '%s': symbol 0x%zx not found for relo #%d\n",
3555                                 relo_sec_name, (size_t)GELF_R_SYM(rel.r_info), i);
3556                         return -LIBBPF_ERRNO__FORMAT;
3557                 }
3558                 if (rel.r_offset % BPF_INSN_SZ) {
3559                         pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n",
3560                                 relo_sec_name, (size_t)GELF_R_SYM(rel.r_info), i);
3561                         return -LIBBPF_ERRNO__FORMAT;
3562                 }
3563
3564                 insn_idx = rel.r_offset / BPF_INSN_SZ;
3565                 /* relocations against static functions are recorded as
3566                  * relocations against the section that contains a function;
3567                  * in such case, symbol will be STT_SECTION and sym.st_name
3568                  * will point to empty string (0), so fetch section name
3569                  * instead
3570                  */
3571                 if (GELF_ST_TYPE(sym.st_info) == STT_SECTION && sym.st_name == 0)
3572                         sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym.st_shndx));
3573                 else
3574                         sym_name = elf_sym_str(obj, sym.st_name);
3575                 sym_name = sym_name ?: "<?";
3576
3577                 pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n",
3578                          relo_sec_name, i, insn_idx, sym_name);
3579
3580                 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
3581                 if (!prog) {
3582                         pr_warn("sec '%s': relo #%d: program not found in section '%s' for insn #%u\n",
3583                                 relo_sec_name, i, sec_name, insn_idx);
3584                         return -LIBBPF_ERRNO__RELOC;
3585                 }
3586
3587                 relos = libbpf_reallocarray(prog->reloc_desc,
3588                                             prog->nr_reloc + 1, sizeof(*relos));
3589                 if (!relos)
3590                         return -ENOMEM;
3591                 prog->reloc_desc = relos;
3592
3593                 /* adjust insn_idx to local BPF program frame of reference */
3594                 insn_idx -= prog->sec_insn_off;
3595                 err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc],
3596                                                 insn_idx, sym_name, &sym, &rel);
3597                 if (err)
3598                         return err;
3599
3600                 prog->nr_reloc++;
3601         }
3602         return 0;
3603 }
3604
3605 static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map)
3606 {
3607         struct bpf_map_def *def = &map->def;
3608         __u32 key_type_id = 0, value_type_id = 0;
3609         int ret;
3610
3611         /* if it's BTF-defined map, we don't need to search for type IDs.
3612          * For struct_ops map, it does not need btf_key_type_id and
3613          * btf_value_type_id.
3614          */
3615         if (map->sec_idx == obj->efile.btf_maps_shndx ||
3616             bpf_map__is_struct_ops(map))
3617                 return 0;
3618
3619         if (!bpf_map__is_internal(map)) {
3620                 ret = btf__get_map_kv_tids(obj->btf, map->name, def->key_size,
3621                                            def->value_size, &key_type_id,
3622                                            &value_type_id);
3623         } else {
3624                 /*
3625                  * LLVM annotates global data differently in BTF, that is,
3626                  * only as '.data', '.bss' or '.rodata'.
3627                  */
3628                 ret = btf__find_by_name(obj->btf,
3629                                 libbpf_type_to_btf_name[map->libbpf_type]);
3630         }
3631         if (ret < 0)
3632                 return ret;
3633
3634         map->btf_key_type_id = key_type_id;
3635         map->btf_value_type_id = bpf_map__is_internal(map) ?
3636                                  ret : value_type_id;
3637         return 0;
3638 }
3639
3640 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
3641 {
3642         struct bpf_map_info info = {};
3643         __u32 len = sizeof(info);
3644         int new_fd, err;
3645         char *new_name;
3646
3647         err = bpf_obj_get_info_by_fd(fd, &info, &len);
3648         if (err)
3649                 return err;
3650
3651         new_name = strdup(info.name);
3652         if (!new_name)
3653                 return -errno;
3654
3655         new_fd = open("/", O_RDONLY | O_CLOEXEC);
3656         if (new_fd < 0) {
3657                 err = -errno;
3658                 goto err_free_new_name;
3659         }
3660
3661         new_fd = dup3(fd, new_fd, O_CLOEXEC);
3662         if (new_fd < 0) {
3663                 err = -errno;
3664                 goto err_close_new_fd;
3665         }
3666
3667         err = zclose(map->fd);
3668         if (err) {
3669                 err = -errno;
3670                 goto err_close_new_fd;
3671         }
3672         free(map->name);
3673
3674         map->fd = new_fd;
3675         map->name = new_name;
3676         map->def.type = info.type;
3677         map->def.key_size = info.key_size;
3678         map->def.value_size = info.value_size;
3679         map->def.max_entries = info.max_entries;
3680         map->def.map_flags = info.map_flags;
3681         map->btf_key_type_id = info.btf_key_type_id;
3682         map->btf_value_type_id = info.btf_value_type_id;
3683         map->reused = true;
3684
3685         return 0;
3686
3687 err_close_new_fd:
3688         close(new_fd);
3689 err_free_new_name:
3690         free(new_name);
3691         return err;
3692 }
3693
3694 __u32 bpf_map__max_entries(const struct bpf_map *map)
3695 {
3696         return map->def.max_entries;
3697 }
3698
3699 int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries)
3700 {
3701         if (map->fd >= 0)
3702                 return -EBUSY;
3703         map->def.max_entries = max_entries;
3704         return 0;
3705 }
3706
3707 int bpf_map__resize(struct bpf_map *map, __u32 max_entries)
3708 {
3709         if (!map || !max_entries)
3710                 return -EINVAL;
3711
3712         return bpf_map__set_max_entries(map, max_entries);
3713 }
3714
3715 static int
3716 bpf_object__probe_loading(struct bpf_object *obj)
3717 {
3718         struct bpf_load_program_attr attr;
3719         char *cp, errmsg[STRERR_BUFSIZE];
3720         struct bpf_insn insns[] = {
3721                 BPF_MOV64_IMM(BPF_REG_0, 0),
3722                 BPF_EXIT_INSN(),
3723         };
3724         int ret;
3725
3726         /* make sure basic loading works */
3727
3728         memset(&attr, 0, sizeof(attr));
3729         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3730         attr.insns = insns;
3731         attr.insns_cnt = ARRAY_SIZE(insns);
3732         attr.license = "GPL";
3733
3734         ret = bpf_load_program_xattr(&attr, NULL, 0);
3735         if (ret < 0) {
3736                 ret = errno;
3737                 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
3738                 pr_warn("Error in %s():%s(%d). Couldn't load trivial BPF "
3739                         "program. Make sure your kernel supports BPF "
3740                         "(CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is "
3741                         "set to big enough value.\n", __func__, cp, ret);
3742                 return -ret;
3743         }
3744         close(ret);
3745
3746         return 0;
3747 }
3748
3749 static int probe_fd(int fd)
3750 {
3751         if (fd >= 0)
3752                 close(fd);
3753         return fd >= 0;
3754 }
3755
3756 static int probe_kern_prog_name(void)
3757 {
3758         struct bpf_load_program_attr attr;
3759         struct bpf_insn insns[] = {
3760                 BPF_MOV64_IMM(BPF_REG_0, 0),
3761                 BPF_EXIT_INSN(),
3762         };
3763         int ret;
3764
3765         /* make sure loading with name works */
3766
3767         memset(&attr, 0, sizeof(attr));
3768         attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3769         attr.insns = insns;
3770         attr.insns_cnt = ARRAY_SIZE(insns);
3771         attr.license = "GPL";
3772         attr.name = "test";
3773         ret = bpf_load_program_xattr(&attr, NULL, 0);
3774         return probe_fd(ret);
3775 }
3776
3777 static int probe_kern_global_data(void)
3778 {
3779         struct bpf_load_program_attr prg_attr;
3780         struct bpf_create_map_attr map_attr;
3781         char *cp, errmsg[STRERR_BUFSIZE];
3782         struct bpf_insn insns[] = {
3783                 BPF_LD_MAP_VALUE(BPF_REG_1, 0, 16),
3784                 BPF_ST_MEM(BPF_DW, BPF_REG_1, 0, 42),
3785                 BPF_MOV64_IMM(BPF_REG_0, 0),
3786                 BPF_EXIT_INSN(),
3787         };
3788         int ret, map;
3789
3790         memset(&map_attr, 0, sizeof(map_attr));
3791         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
3792         map_attr.key_size = sizeof(int);
3793         map_attr.value_size = 32;
3794         map_attr.max_entries = 1;
3795
3796         map = bpf_create_map_xattr(&map_attr);
3797         if (map < 0) {
3798                 ret = -errno;
3799                 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
3800                 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
3801                         __func__, cp, -ret);
3802                 return ret;
3803         }
3804
3805         insns[0].imm = map;
3806
3807         memset(&prg_attr, 0, sizeof(prg_attr));
3808         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3809         prg_attr.insns = insns;
3810         prg_attr.insns_cnt = ARRAY_SIZE(insns);
3811         prg_attr.license = "GPL";
3812
3813         ret = bpf_load_program_xattr(&prg_attr, NULL, 0);
3814         close(map);
3815         return probe_fd(ret);
3816 }
3817
3818 static int probe_kern_btf(void)
3819 {
3820         static const char strs[] = "\0int";
3821         __u32 types[] = {
3822                 /* int */
3823                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
3824         };
3825
3826         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
3827                                              strs, sizeof(strs)));
3828 }
3829
3830 static int probe_kern_btf_func(void)
3831 {
3832         static const char strs[] = "\0int\0x\0a";
3833         /* void x(int a) {} */
3834         __u32 types[] = {
3835                 /* int */
3836                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
3837                 /* FUNC_PROTO */                                /* [2] */
3838                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
3839                 BTF_PARAM_ENC(7, 1),
3840                 /* FUNC x */                                    /* [3] */
3841                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0), 2),
3842         };
3843
3844         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
3845                                              strs, sizeof(strs)));
3846 }
3847
3848 static int probe_kern_btf_func_global(void)
3849 {
3850         static const char strs[] = "\0int\0x\0a";
3851         /* static void x(int a) {} */
3852         __u32 types[] = {
3853                 /* int */
3854                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
3855                 /* FUNC_PROTO */                                /* [2] */
3856                 BTF_TYPE_ENC(0, BTF_INFO_ENC(BTF_KIND_FUNC_PROTO, 0, 1), 0),
3857                 BTF_PARAM_ENC(7, 1),
3858                 /* FUNC x BTF_FUNC_GLOBAL */                    /* [3] */
3859                 BTF_TYPE_ENC(5, BTF_INFO_ENC(BTF_KIND_FUNC, 0, BTF_FUNC_GLOBAL), 2),
3860         };
3861
3862         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
3863                                              strs, sizeof(strs)));
3864 }
3865
3866 static int probe_kern_btf_datasec(void)
3867 {
3868         static const char strs[] = "\0x\0.data";
3869         /* static int a; */
3870         __u32 types[] = {
3871                 /* int */
3872                 BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 32, 4),  /* [1] */
3873                 /* VAR x */                                     /* [2] */
3874                 BTF_TYPE_ENC(1, BTF_INFO_ENC(BTF_KIND_VAR, 0, 0), 1),
3875                 BTF_VAR_STATIC,
3876                 /* DATASEC val */                               /* [3] */
3877                 BTF_TYPE_ENC(3, BTF_INFO_ENC(BTF_KIND_DATASEC, 0, 1), 4),
3878                 BTF_VAR_SECINFO_ENC(2, 0, 4),
3879         };
3880
3881         return probe_fd(libbpf__load_raw_btf((char *)types, sizeof(types),
3882                                              strs, sizeof(strs)));
3883 }
3884
3885 static int probe_kern_array_mmap(void)
3886 {
3887         struct bpf_create_map_attr attr = {
3888                 .map_type = BPF_MAP_TYPE_ARRAY,
3889                 .map_flags = BPF_F_MMAPABLE,
3890                 .key_size = sizeof(int),
3891                 .value_size = sizeof(int),
3892                 .max_entries = 1,
3893         };
3894
3895         return probe_fd(bpf_create_map_xattr(&attr));
3896 }
3897
3898 static int probe_kern_exp_attach_type(void)
3899 {
3900         struct bpf_load_program_attr attr;
3901         struct bpf_insn insns[] = {
3902                 BPF_MOV64_IMM(BPF_REG_0, 0),
3903                 BPF_EXIT_INSN(),
3904         };
3905
3906         memset(&attr, 0, sizeof(attr));
3907         /* use any valid combination of program type and (optional)
3908          * non-zero expected attach type (i.e., not a BPF_CGROUP_INET_INGRESS)
3909          * to see if kernel supports expected_attach_type field for
3910          * BPF_PROG_LOAD command
3911          */
3912         attr.prog_type = BPF_PROG_TYPE_CGROUP_SOCK;
3913         attr.expected_attach_type = BPF_CGROUP_INET_SOCK_CREATE;
3914         attr.insns = insns;
3915         attr.insns_cnt = ARRAY_SIZE(insns);
3916         attr.license = "GPL";
3917
3918         return probe_fd(bpf_load_program_xattr(&attr, NULL, 0));
3919 }
3920
3921 static int probe_kern_probe_read_kernel(void)
3922 {
3923         struct bpf_load_program_attr attr;
3924         struct bpf_insn insns[] = {
3925                 BPF_MOV64_REG(BPF_REG_1, BPF_REG_10),   /* r1 = r10 (fp) */
3926                 BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8),  /* r1 += -8 */
3927                 BPF_MOV64_IMM(BPF_REG_2, 8),            /* r2 = 8 */
3928                 BPF_MOV64_IMM(BPF_REG_3, 0),            /* r3 = 0 */
3929                 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel),
3930                 BPF_EXIT_INSN(),
3931         };
3932
3933         memset(&attr, 0, sizeof(attr));
3934         attr.prog_type = BPF_PROG_TYPE_KPROBE;
3935         attr.insns = insns;
3936         attr.insns_cnt = ARRAY_SIZE(insns);
3937         attr.license = "GPL";
3938
3939         return probe_fd(bpf_load_program_xattr(&attr, NULL, 0));
3940 }
3941
3942 static int probe_prog_bind_map(void)
3943 {
3944         struct bpf_load_program_attr prg_attr;
3945         struct bpf_create_map_attr map_attr;
3946         char *cp, errmsg[STRERR_BUFSIZE];
3947         struct bpf_insn insns[] = {
3948                 BPF_MOV64_IMM(BPF_REG_0, 0),
3949                 BPF_EXIT_INSN(),
3950         };
3951         int ret, map, prog;
3952
3953         memset(&map_attr, 0, sizeof(map_attr));
3954         map_attr.map_type = BPF_MAP_TYPE_ARRAY;
3955         map_attr.key_size = sizeof(int);
3956         map_attr.value_size = 32;
3957         map_attr.max_entries = 1;
3958
3959         map = bpf_create_map_xattr(&map_attr);
3960         if (map < 0) {
3961                 ret = -errno;
3962                 cp = libbpf_strerror_r(ret, errmsg, sizeof(errmsg));
3963                 pr_warn("Error in %s():%s(%d). Couldn't create simple array map.\n",
3964                         __func__, cp, -ret);
3965                 return ret;
3966         }
3967
3968         memset(&prg_attr, 0, sizeof(prg_attr));
3969         prg_attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
3970         prg_attr.insns = insns;
3971         prg_attr.insns_cnt = ARRAY_SIZE(insns);
3972         prg_attr.license = "GPL";
3973
3974         prog = bpf_load_program_xattr(&prg_attr, NULL, 0);
3975         if (prog < 0) {
3976                 close(map);
3977                 return 0;
3978         }
3979
3980         ret = bpf_prog_bind_map(prog, map, NULL);
3981
3982         close(map);
3983         close(prog);
3984
3985         return ret >= 0;
3986 }
3987
3988 static int probe_module_btf(void)
3989 {
3990         static const char strs[] = "\0int";
3991         __u32 types[] = {
3992                 /* int */
3993                 BTF_TYPE_INT_ENC(1, BTF_INT_SIGNED, 0, 32, 4),
3994         };
3995         struct bpf_btf_info info;
3996         __u32 len = sizeof(info);
3997         char name[16];
3998         int fd, err;
3999
4000         fd = libbpf__load_raw_btf((char *)types, sizeof(types), strs, sizeof(strs));
4001         if (fd < 0)
4002                 return 0; /* BTF not supported at all */
4003
4004         memset(&info, 0, sizeof(info));
4005         info.name = ptr_to_u64(name);
4006         info.name_len = sizeof(name);
4007
4008         /* check that BPF_OBJ_GET_INFO_BY_FD supports specifying name pointer;
4009          * kernel's module BTF support coincides with support for
4010          * name/name_len fields in struct bpf_btf_info.
4011          */
4012         err = bpf_obj_get_info_by_fd(fd, &info, &len);
4013         close(fd);
4014         return !err;
4015 }
4016
4017 enum kern_feature_result {
4018         FEAT_UNKNOWN = 0,
4019         FEAT_SUPPORTED = 1,
4020         FEAT_MISSING = 2,
4021 };
4022
4023 typedef int (*feature_probe_fn)(void);
4024
4025 static struct kern_feature_desc {
4026         const char *desc;
4027         feature_probe_fn probe;
4028         enum kern_feature_result res;
4029 } feature_probes[__FEAT_CNT] = {
4030         [FEAT_PROG_NAME] = {
4031                 "BPF program name", probe_kern_prog_name,
4032         },
4033         [FEAT_GLOBAL_DATA] = {
4034                 "global variables", probe_kern_global_data,
4035         },
4036         [FEAT_BTF] = {
4037                 "minimal BTF", probe_kern_btf,
4038         },
4039         [FEAT_BTF_FUNC] = {
4040                 "BTF functions", probe_kern_btf_func,
4041         },
4042         [FEAT_BTF_GLOBAL_FUNC] = {
4043                 "BTF global function", probe_kern_btf_func_global,
4044         },
4045         [FEAT_BTF_DATASEC] = {
4046                 "BTF data section and variable", probe_kern_btf_datasec,
4047         },
4048         [FEAT_ARRAY_MMAP] = {
4049                 "ARRAY map mmap()", probe_kern_array_mmap,
4050         },
4051         [FEAT_EXP_ATTACH_TYPE] = {
4052                 "BPF_PROG_LOAD expected_attach_type attribute",
4053                 probe_kern_exp_attach_type,
4054         },
4055         [FEAT_PROBE_READ_KERN] = {
4056                 "bpf_probe_read_kernel() helper", probe_kern_probe_read_kernel,
4057         },
4058         [FEAT_PROG_BIND_MAP] = {
4059                 "BPF_PROG_BIND_MAP support", probe_prog_bind_map,
4060         },
4061         [FEAT_MODULE_BTF] = {
4062                 "module BTF support", probe_module_btf,
4063         },
4064 };
4065
4066 static bool kernel_supports(enum kern_feature_id feat_id)
4067 {
4068         struct kern_feature_desc *feat = &feature_probes[feat_id];
4069         int ret;
4070
4071         if (READ_ONCE(feat->res) == FEAT_UNKNOWN) {
4072                 ret = feat->probe();
4073                 if (ret > 0) {
4074                         WRITE_ONCE(feat->res, FEAT_SUPPORTED);
4075                 } else if (ret == 0) {
4076                         WRITE_ONCE(feat->res, FEAT_MISSING);
4077                 } else {
4078                         pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret);
4079                         WRITE_ONCE(feat->res, FEAT_MISSING);
4080                 }
4081         }
4082
4083         return READ_ONCE(feat->res) == FEAT_SUPPORTED;
4084 }
4085
4086 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
4087 {
4088         struct bpf_map_info map_info = {};
4089         char msg[STRERR_BUFSIZE];
4090         __u32 map_info_len;
4091
4092         map_info_len = sizeof(map_info);
4093
4094         if (bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len)) {
4095                 pr_warn("failed to get map info for map FD %d: %s\n",
4096                         map_fd, libbpf_strerror_r(errno, msg, sizeof(msg)));
4097                 return false;
4098         }
4099
4100         return (map_info.type == map->def.type &&
4101                 map_info.key_size == map->def.key_size &&
4102                 map_info.value_size == map->def.value_size &&
4103                 map_info.max_entries == map->def.max_entries &&
4104                 map_info.map_flags == map->def.map_flags);
4105 }
4106
4107 static int
4108 bpf_object__reuse_map(struct bpf_map *map)
4109 {
4110         char *cp, errmsg[STRERR_BUFSIZE];
4111         int err, pin_fd;
4112
4113         pin_fd = bpf_obj_get(map->pin_path);
4114         if (pin_fd < 0) {
4115                 err = -errno;
4116                 if (err == -ENOENT) {
4117                         pr_debug("found no pinned map to reuse at '%s'\n",
4118                                  map->pin_path);
4119                         return 0;
4120                 }
4121
4122                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
4123                 pr_warn("couldn't retrieve pinned map '%s': %s\n",
4124                         map->pin_path, cp);
4125                 return err;
4126         }
4127
4128         if (!map_is_reuse_compat(map, pin_fd)) {
4129                 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
4130                         map->pin_path);
4131                 close(pin_fd);
4132                 return -EINVAL;
4133         }
4134
4135         err = bpf_map__reuse_fd(map, pin_fd);
4136         if (err) {
4137                 close(pin_fd);
4138                 return err;
4139         }
4140         map->pinned = true;
4141         pr_debug("reused pinned map at '%s'\n", map->pin_path);
4142
4143         return 0;
4144 }
4145
4146 static int
4147 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
4148 {
4149         enum libbpf_map_type map_type = map->libbpf_type;
4150         char *cp, errmsg[STRERR_BUFSIZE];
4151         int err, zero = 0;
4152
4153         err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
4154         if (err) {
4155                 err = -errno;
4156                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4157                 pr_warn("Error setting initial map(%s) contents: %s\n",
4158                         map->name, cp);
4159                 return err;
4160         }
4161
4162         /* Freeze .rodata and .kconfig map as read-only from syscall side. */
4163         if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
4164                 err = bpf_map_freeze(map->fd);
4165                 if (err) {
4166                         err = -errno;
4167                         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4168                         pr_warn("Error freezing map(%s) as read-only: %s\n",
4169                                 map->name, cp);
4170                         return err;
4171                 }
4172         }
4173         return 0;
4174 }
4175
4176 static void bpf_map__destroy(struct bpf_map *map);
4177
4178 static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map)
4179 {
4180         struct bpf_create_map_attr create_attr;
4181         struct bpf_map_def *def = &map->def;
4182
4183         memset(&create_attr, 0, sizeof(create_attr));
4184
4185         if (kernel_supports(FEAT_PROG_NAME))
4186                 create_attr.name = map->name;
4187         create_attr.map_ifindex = map->map_ifindex;
4188         create_attr.map_type = def->type;
4189         create_attr.map_flags = def->map_flags;
4190         create_attr.key_size = def->key_size;
4191         create_attr.value_size = def->value_size;
4192         create_attr.numa_node = map->numa_node;
4193
4194         if (def->type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !def->max_entries) {
4195                 int nr_cpus;
4196
4197                 nr_cpus = libbpf_num_possible_cpus();
4198                 if (nr_cpus < 0) {
4199                         pr_warn("map '%s': failed to determine number of system CPUs: %d\n",
4200                                 map->name, nr_cpus);
4201                         return nr_cpus;
4202                 }
4203                 pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus);
4204                 create_attr.max_entries = nr_cpus;
4205         } else {
4206                 create_attr.max_entries = def->max_entries;
4207         }
4208
4209         if (bpf_map__is_struct_ops(map))
4210                 create_attr.btf_vmlinux_value_type_id =
4211                         map->btf_vmlinux_value_type_id;
4212
4213         create_attr.btf_fd = 0;
4214         create_attr.btf_key_type_id = 0;
4215         create_attr.btf_value_type_id = 0;
4216         if (obj->btf && btf__fd(obj->btf) >= 0 && !bpf_map_find_btf_info(obj, map)) {
4217                 create_attr.btf_fd = btf__fd(obj->btf);
4218                 create_attr.btf_key_type_id = map->btf_key_type_id;
4219                 create_attr.btf_value_type_id = map->btf_value_type_id;
4220         }
4221
4222         if (bpf_map_type__is_map_in_map(def->type)) {
4223                 if (map->inner_map) {
4224                         int err;
4225
4226                         err = bpf_object__create_map(obj, map->inner_map);
4227                         if (err) {
4228                                 pr_warn("map '%s': failed to create inner map: %d\n",
4229                                         map->name, err);
4230                                 return err;
4231                         }
4232                         map->inner_map_fd = bpf_map__fd(map->inner_map);
4233                 }
4234                 if (map->inner_map_fd >= 0)
4235                         create_attr.inner_map_fd = map->inner_map_fd;
4236         }
4237
4238         map->fd = bpf_create_map_xattr(&create_attr);
4239         if (map->fd < 0 && (create_attr.btf_key_type_id ||
4240                             create_attr.btf_value_type_id)) {
4241                 char *cp, errmsg[STRERR_BUFSIZE];
4242                 int err = -errno;
4243
4244                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4245                 pr_warn("Error in bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n",
4246                         map->name, cp, err);
4247                 create_attr.btf_fd = 0;
4248                 create_attr.btf_key_type_id = 0;
4249                 create_attr.btf_value_type_id = 0;
4250                 map->btf_key_type_id = 0;
4251                 map->btf_value_type_id = 0;
4252                 map->fd = bpf_create_map_xattr(&create_attr);
4253         }
4254
4255         if (map->fd < 0)
4256                 return -errno;
4257
4258         if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) {
4259                 bpf_map__destroy(map->inner_map);
4260                 zfree(&map->inner_map);
4261         }
4262
4263         return 0;
4264 }
4265
4266 static int init_map_slots(struct bpf_map *map)
4267 {
4268         const struct bpf_map *targ_map;
4269         unsigned int i;
4270         int fd, err;
4271
4272         for (i = 0; i < map->init_slots_sz; i++) {
4273                 if (!map->init_slots[i])
4274                         continue;
4275
4276                 targ_map = map->init_slots[i];
4277                 fd = bpf_map__fd(targ_map);
4278                 err = bpf_map_update_elem(map->fd, &i, &fd, 0);
4279                 if (err) {
4280                         err = -errno;
4281                         pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %d\n",
4282                                 map->name, i, targ_map->name,
4283                                 fd, err);
4284                         return err;
4285                 }
4286                 pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n",
4287                          map->name, i, targ_map->name, fd);
4288         }
4289
4290         zfree(&map->init_slots);
4291         map->init_slots_sz = 0;
4292
4293         return 0;
4294 }
4295
4296 static int
4297 bpf_object__create_maps(struct bpf_object *obj)
4298 {
4299         struct bpf_map *map;
4300         char *cp, errmsg[STRERR_BUFSIZE];
4301         unsigned int i, j;
4302         int err;
4303
4304         for (i = 0; i < obj->nr_maps; i++) {
4305                 map = &obj->maps[i];
4306
4307                 if (map->pin_path) {
4308                         err = bpf_object__reuse_map(map);
4309                         if (err) {
4310                                 pr_warn("map '%s': error reusing pinned map\n",
4311                                         map->name);
4312                                 goto err_out;
4313                         }
4314                 }
4315
4316                 if (map->fd >= 0) {
4317                         pr_debug("map '%s': skipping creation (preset fd=%d)\n",
4318                                  map->name, map->fd);
4319                 } else {
4320                         err = bpf_object__create_map(obj, map);
4321                         if (err)
4322                                 goto err_out;
4323
4324                         pr_debug("map '%s': created successfully, fd=%d\n",
4325                                  map->name, map->fd);
4326
4327                         if (bpf_map__is_internal(map)) {
4328                                 err = bpf_object__populate_internal_map(obj, map);
4329                                 if (err < 0) {
4330                                         zclose(map->fd);
4331                                         goto err_out;
4332                                 }
4333                         }
4334
4335                         if (map->init_slots_sz) {
4336                                 err = init_map_slots(map);
4337                                 if (err < 0) {
4338                                         zclose(map->fd);
4339                                         goto err_out;
4340                                 }
4341                         }
4342                 }
4343
4344                 if (map->pin_path && !map->pinned) {
4345                         err = bpf_map__pin(map, NULL);
4346                         if (err) {
4347                                 pr_warn("map '%s': failed to auto-pin at '%s': %d\n",
4348                                         map->name, map->pin_path, err);
4349                                 zclose(map->fd);
4350                                 goto err_out;
4351                         }
4352                 }
4353         }
4354
4355         return 0;
4356
4357 err_out:
4358         cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
4359         pr_warn("map '%s': failed to create: %s(%d)\n", map->name, cp, err);
4360         pr_perm_msg(err);
4361         for (j = 0; j < i; j++)
4362                 zclose(obj->maps[j].fd);
4363         return err;
4364 }
4365
4366 #define BPF_CORE_SPEC_MAX_LEN 64
4367
4368 /* represents BPF CO-RE field or array element accessor */
4369 struct bpf_core_accessor {
4370         __u32 type_id;          /* struct/union type or array element type */
4371         __u32 idx;              /* field index or array index */
4372         const char *name;       /* field name or NULL for array accessor */
4373 };
4374
4375 struct bpf_core_spec {
4376         const struct btf *btf;
4377         /* high-level spec: named fields and array indices only */
4378         struct bpf_core_accessor spec[BPF_CORE_SPEC_MAX_LEN];
4379         /* original unresolved (no skip_mods_or_typedefs) root type ID */
4380         __u32 root_type_id;
4381         /* CO-RE relocation kind */
4382         enum bpf_core_relo_kind relo_kind;
4383         /* high-level spec length */
4384         int len;
4385         /* raw, low-level spec: 1-to-1 with accessor spec string */
4386         int raw_spec[BPF_CORE_SPEC_MAX_LEN];
4387         /* raw spec length */
4388         int raw_len;
4389         /* field bit offset represented by spec */
4390         __u32 bit_offset;
4391 };
4392
4393 static bool str_is_empty(const char *s)
4394 {
4395         return !s || !s[0];
4396 }
4397
4398 static bool is_flex_arr(const struct btf *btf,
4399                         const struct bpf_core_accessor *acc,
4400                         const struct btf_array *arr)
4401 {
4402         const struct btf_type *t;
4403
4404         /* not a flexible array, if not inside a struct or has non-zero size */
4405         if (!acc->name || arr->nelems > 0)
4406                 return false;
4407
4408         /* has to be the last member of enclosing struct */
4409         t = btf__type_by_id(btf, acc->type_id);
4410         return acc->idx == btf_vlen(t) - 1;
4411 }
4412
4413 static const char *core_relo_kind_str(enum bpf_core_relo_kind kind)
4414 {
4415         switch (kind) {
4416         case BPF_FIELD_BYTE_OFFSET: return "byte_off";
4417         case BPF_FIELD_BYTE_SIZE: return "byte_sz";
4418         case BPF_FIELD_EXISTS: return "field_exists";
4419         case BPF_FIELD_SIGNED: return "signed";
4420         case BPF_FIELD_LSHIFT_U64: return "lshift_u64";
4421         case BPF_FIELD_RSHIFT_U64: return "rshift_u64";
4422         case BPF_TYPE_ID_LOCAL: return "local_type_id";
4423         case BPF_TYPE_ID_TARGET: return "target_type_id";
4424         case BPF_TYPE_EXISTS: return "type_exists";
4425         case BPF_TYPE_SIZE: return "type_size";
4426         case BPF_ENUMVAL_EXISTS: return "enumval_exists";
4427         case BPF_ENUMVAL_VALUE: return "enumval_value";
4428         default: return "unknown";
4429         }
4430 }
4431
4432 static bool core_relo_is_field_based(enum bpf_core_relo_kind kind)
4433 {
4434         switch (kind) {
4435         case BPF_FIELD_BYTE_OFFSET:
4436         case BPF_FIELD_BYTE_SIZE:
4437         case BPF_FIELD_EXISTS:
4438         case BPF_FIELD_SIGNED:
4439         case BPF_FIELD_LSHIFT_U64:
4440         case BPF_FIELD_RSHIFT_U64:
4441                 return true;
4442         default:
4443                 return false;
4444         }
4445 }
4446
4447 static bool core_relo_is_type_based(enum bpf_core_relo_kind kind)
4448 {
4449         switch (kind) {
4450         case BPF_TYPE_ID_LOCAL:
4451         case BPF_TYPE_ID_TARGET:
4452         case BPF_TYPE_EXISTS:
4453         case BPF_TYPE_SIZE:
4454                 return true;
4455         default:
4456                 return false;
4457         }
4458 }
4459
4460 static bool core_relo_is_enumval_based(enum bpf_core_relo_kind kind)
4461 {
4462         switch (kind) {
4463         case BPF_ENUMVAL_EXISTS:
4464         case BPF_ENUMVAL_VALUE:
4465                 return true;
4466         default:
4467                 return false;
4468         }
4469 }
4470
4471 /*
4472  * Turn bpf_core_relo into a low- and high-level spec representation,
4473  * validating correctness along the way, as well as calculating resulting
4474  * field bit offset, specified by accessor string. Low-level spec captures
4475  * every single level of nestedness, including traversing anonymous
4476  * struct/union members. High-level one only captures semantically meaningful
4477  * "turning points": named fields and array indicies.
4478  * E.g., for this case:
4479  *
4480  *   struct sample {
4481  *       int __unimportant;
4482  *       struct {
4483  *           int __1;
4484  *           int __2;
4485  *           int a[7];
4486  *       };
4487  *   };
4488  *
4489  *   struct sample *s = ...;
4490  *
4491  *   int x = &s->a[3]; // access string = '0:1:2:3'
4492  *
4493  * Low-level spec has 1:1 mapping with each element of access string (it's
4494  * just a parsed access string representation): [0, 1, 2, 3].
4495  *
4496  * High-level spec will capture only 3 points:
4497  *   - intial zero-index access by pointer (&s->... is the same as &s[0]...);
4498  *   - field 'a' access (corresponds to '2' in low-level spec);
4499  *   - array element #3 access (corresponds to '3' in low-level spec).
4500  *
4501  * Type-based relocations (TYPE_EXISTS/TYPE_SIZE,
4502  * TYPE_ID_LOCAL/TYPE_ID_TARGET) don't capture any field information. Their
4503  * spec and raw_spec are kept empty.
4504  *
4505  * Enum value-based relocations (ENUMVAL_EXISTS/ENUMVAL_VALUE) use access
4506  * string to specify enumerator's value index that need to be relocated.
4507  */
4508 static int bpf_core_parse_spec(const struct btf *btf,
4509                                __u32 type_id,
4510                                const char *spec_str,
4511                                enum bpf_core_relo_kind relo_kind,
4512                                struct bpf_core_spec *spec)
4513 {
4514         int access_idx, parsed_len, i;
4515         struct bpf_core_accessor *acc;
4516         const struct btf_type *t;
4517         const char *name;
4518         __u32 id;
4519         __s64 sz;
4520
4521         if (str_is_empty(spec_str) || *spec_str == ':')
4522                 return -EINVAL;
4523
4524         memset(spec, 0, sizeof(*spec));
4525         spec->btf = btf;
4526         spec->root_type_id = type_id;
4527         spec->relo_kind = relo_kind;
4528
4529         /* type-based relocations don't have a field access string */
4530         if (core_relo_is_type_based(relo_kind)) {
4531                 if (strcmp(spec_str, "0"))
4532                         return -EINVAL;
4533                 return 0;
4534         }
4535
4536         /* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
4537         while (*spec_str) {
4538                 if (*spec_str == ':')
4539                         ++spec_str;
4540                 if (sscanf(spec_str, "%d%n", &access_idx, &parsed_len) != 1)
4541                         return -EINVAL;
4542                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
4543                         return -E2BIG;
4544                 spec_str += parsed_len;
4545                 spec->raw_spec[spec->raw_len++] = access_idx;
4546         }
4547
4548         if (spec->raw_len == 0)
4549                 return -EINVAL;
4550
4551         t = skip_mods_and_typedefs(btf, type_id, &id);
4552         if (!t)
4553                 return -EINVAL;
4554
4555         access_idx = spec->raw_spec[0];
4556         acc = &spec->spec[0];
4557         acc->type_id = id;
4558         acc->idx = access_idx;
4559         spec->len++;
4560
4561         if (core_relo_is_enumval_based(relo_kind)) {
4562                 if (!btf_is_enum(t) || spec->raw_len > 1 || access_idx >= btf_vlen(t))
4563                         return -EINVAL;
4564
4565                 /* record enumerator name in a first accessor */
4566                 acc->name = btf__name_by_offset(btf, btf_enum(t)[access_idx].name_off);
4567                 return 0;
4568         }
4569
4570         if (!core_relo_is_field_based(relo_kind))
4571                 return -EINVAL;
4572
4573         sz = btf__resolve_size(btf, id);
4574         if (sz < 0)
4575                 return sz;
4576         spec->bit_offset = access_idx * sz * 8;
4577
4578         for (i = 1; i < spec->raw_len; i++) {
4579                 t = skip_mods_and_typedefs(btf, id, &id);
4580                 if (!t)
4581                         return -EINVAL;
4582
4583                 access_idx = spec->raw_spec[i];
4584                 acc = &spec->spec[spec->len];
4585
4586                 if (btf_is_composite(t)) {
4587                         const struct btf_member *m;
4588                         __u32 bit_offset;
4589
4590                         if (access_idx >= btf_vlen(t))
4591                                 return -EINVAL;
4592
4593                         bit_offset = btf_member_bit_offset(t, access_idx);
4594                         spec->bit_offset += bit_offset;
4595
4596                         m = btf_members(t) + access_idx;
4597                         if (m->name_off) {
4598                                 name = btf__name_by_offset(btf, m->name_off);
4599                                 if (str_is_empty(name))
4600                                         return -EINVAL;
4601
4602                                 acc->type_id = id;
4603                                 acc->idx = access_idx;
4604                                 acc->name = name;
4605                                 spec->len++;
4606                         }
4607
4608                         id = m->type;
4609                 } else if (btf_is_array(t)) {
4610                         const struct btf_array *a = btf_array(t);
4611                         bool flex;
4612
4613                         t = skip_mods_and_typedefs(btf, a->type, &id);
4614                         if (!t)
4615                                 return -EINVAL;
4616
4617                         flex = is_flex_arr(btf, acc - 1, a);
4618                         if (!flex && access_idx >= a->nelems)
4619                                 return -EINVAL;
4620
4621                         spec->spec[spec->len].type_id = id;
4622                         spec->spec[spec->len].idx = access_idx;
4623                         spec->len++;
4624
4625                         sz = btf__resolve_size(btf, id);
4626                         if (sz < 0)
4627                                 return sz;
4628                         spec->bit_offset += access_idx * sz * 8;
4629                 } else {
4630                         pr_warn("relo for [%u] %s (at idx %d) captures type [%d] of unexpected kind %s\n",
4631                                 type_id, spec_str, i, id, btf_kind_str(t));
4632                         return -EINVAL;
4633                 }
4634         }
4635
4636         return 0;
4637 }
4638
4639 static bool bpf_core_is_flavor_sep(const char *s)
4640 {
4641         /* check X___Y name pattern, where X and Y are not underscores */
4642         return s[0] != '_' &&                                 /* X */
4643                s[1] == '_' && s[2] == '_' && s[3] == '_' &&   /* ___ */
4644                s[4] != '_';                                   /* Y */
4645 }
4646
4647 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
4648  * before last triple underscore. Struct name part after last triple
4649  * underscore is ignored by BPF CO-RE relocation during relocation matching.
4650  */
4651 static size_t bpf_core_essential_name_len(const char *name)
4652 {
4653         size_t n = strlen(name);
4654         int i;
4655
4656         for (i = n - 5; i >= 0; i--) {
4657                 if (bpf_core_is_flavor_sep(name + i))
4658                         return i + 1;
4659         }
4660         return n;
4661 }
4662
4663 struct core_cand
4664 {
4665         const struct btf *btf;
4666         const struct btf_type *t;
4667         const char *name;
4668         __u32 id;
4669 };
4670
4671 /* dynamically sized list of type IDs and its associated struct btf */
4672 struct core_cand_list {
4673         struct core_cand *cands;
4674         int len;
4675 };
4676
4677 static void bpf_core_free_cands(struct core_cand_list *cands)
4678 {
4679         free(cands->cands);
4680         free(cands);
4681 }
4682
4683 static int bpf_core_add_cands(struct core_cand *local_cand,
4684                               size_t local_essent_len,
4685                               const struct btf *targ_btf,
4686                               const char *targ_btf_name,
4687                               int targ_start_id,
4688                               struct core_cand_list *cands)
4689 {
4690         struct core_cand *new_cands, *cand;
4691         const struct btf_type *t;
4692         const char *targ_name;
4693         size_t targ_essent_len;
4694         int n, i;
4695
4696         n = btf__get_nr_types(targ_btf);
4697         for (i = targ_start_id; i <= n; i++) {
4698                 t = btf__type_by_id(targ_btf, i);
4699                 if (btf_kind(t) != btf_kind(local_cand->t))
4700                         continue;
4701
4702                 targ_name = btf__name_by_offset(targ_btf, t->name_off);
4703                 if (str_is_empty(targ_name))
4704                         continue;
4705
4706                 targ_essent_len = bpf_core_essential_name_len(targ_name);
4707                 if (targ_essent_len != local_essent_len)
4708                         continue;
4709
4710                 if (strncmp(local_cand->name, targ_name, local_essent_len) != 0)
4711                         continue;
4712
4713                 pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n",
4714                          local_cand->id, btf_kind_str(local_cand->t),
4715                          local_cand->name, i, btf_kind_str(t), targ_name,
4716                          targ_btf_name);
4717                 new_cands = libbpf_reallocarray(cands->cands, cands->len + 1,
4718                                               sizeof(*cands->cands));
4719                 if (!new_cands)
4720                         return -ENOMEM;
4721
4722                 cand = &new_cands[cands->len];
4723                 cand->btf = targ_btf;
4724                 cand->t = t;
4725                 cand->name = targ_name;
4726                 cand->id = i;
4727
4728                 cands->cands = new_cands;
4729                 cands->len++;
4730         }
4731         return 0;
4732 }
4733
4734 static int load_module_btfs(struct bpf_object *obj)
4735 {
4736         struct bpf_btf_info info;
4737         struct module_btf *mod_btf;
4738         struct btf *btf;
4739         char name[64];
4740         __u32 id = 0, len;
4741         int err, fd;
4742
4743         if (obj->btf_modules_loaded)
4744                 return 0;
4745
4746         /* don't do this again, even if we find no module BTFs */
4747         obj->btf_modules_loaded = true;
4748
4749         /* kernel too old to support module BTFs */
4750         if (!kernel_supports(FEAT_MODULE_BTF))
4751                 return 0;
4752
4753         while (true) {
4754                 err = bpf_btf_get_next_id(id, &id);
4755                 if (err && errno == ENOENT)
4756                         return 0;
4757                 if (err) {
4758                         err = -errno;
4759                         pr_warn("failed to iterate BTF objects: %d\n", err);
4760                         return err;
4761                 }
4762
4763                 fd = bpf_btf_get_fd_by_id(id);
4764                 if (fd < 0) {
4765                         if (errno == ENOENT)
4766                                 continue; /* expected race: BTF was unloaded */
4767                         err = -errno;
4768                         pr_warn("failed to get BTF object #%d FD: %d\n", id, err);
4769                         return err;
4770                 }
4771
4772                 len = sizeof(info);
4773                 memset(&info, 0, sizeof(info));
4774                 info.name = ptr_to_u64(name);
4775                 info.name_len = sizeof(name);
4776
4777                 err = bpf_obj_get_info_by_fd(fd, &info, &len);
4778                 if (err) {
4779                         err = -errno;
4780                         pr_warn("failed to get BTF object #%d info: %d\n", id, err);
4781                         goto err_out;
4782                 }
4783
4784                 /* ignore non-module BTFs */
4785                 if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) {
4786                         close(fd);
4787                         continue;
4788                 }
4789
4790                 btf = btf_get_from_fd(fd, obj->btf_vmlinux);
4791                 if (IS_ERR(btf)) {
4792                         pr_warn("failed to load module [%s]'s BTF object #%d: %ld\n",
4793                                 name, id, PTR_ERR(btf));
4794                         err = PTR_ERR(btf);
4795                         goto err_out;
4796                 }
4797
4798                 err = btf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap,
4799                                      sizeof(*obj->btf_modules), obj->btf_module_cnt + 1);
4800                 if (err)
4801                         goto err_out;
4802
4803                 mod_btf = &obj->btf_modules[obj->btf_module_cnt++];
4804
4805                 mod_btf->btf = btf;
4806                 mod_btf->id = id;
4807                 mod_btf->fd = fd;
4808                 mod_btf->name = strdup(name);
4809                 if (!mod_btf->name) {
4810                         err = -ENOMEM;
4811                         goto err_out;
4812                 }
4813                 continue;
4814
4815 err_out:
4816                 close(fd);
4817                 return err;
4818         }
4819
4820         return 0;
4821 }
4822
4823 static struct core_cand_list *
4824 bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id)
4825 {
4826         struct core_cand local_cand = {};
4827         struct core_cand_list *cands;
4828         const struct btf *main_btf;
4829         size_t local_essent_len;
4830         int err, i;
4831
4832         local_cand.btf = local_btf;
4833         local_cand.t = btf__type_by_id(local_btf, local_type_id);
4834         if (!local_cand.t)
4835                 return ERR_PTR(-EINVAL);
4836
4837         local_cand.name = btf__name_by_offset(local_btf, local_cand.t->name_off);
4838         if (str_is_empty(local_cand.name))
4839                 return ERR_PTR(-EINVAL);
4840         local_essent_len = bpf_core_essential_name_len(local_cand.name);
4841
4842         cands = calloc(1, sizeof(*cands));
4843         if (!cands)
4844                 return ERR_PTR(-ENOMEM);
4845
4846         /* Attempt to find target candidates in vmlinux BTF first */
4847         main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux;
4848         err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands);
4849         if (err)
4850                 goto err_out;
4851
4852         /* if vmlinux BTF has any candidate, don't got for module BTFs */
4853         if (cands->len)
4854                 return cands;
4855
4856         /* if vmlinux BTF was overridden, don't attempt to load module BTFs */
4857         if (obj->btf_vmlinux_override)
4858                 return cands;
4859
4860         /* now look through module BTFs, trying to still find candidates */
4861         err = load_module_btfs(obj);
4862         if (err)
4863                 goto err_out;
4864
4865         for (i = 0; i < obj->btf_module_cnt; i++) {
4866                 err = bpf_core_add_cands(&local_cand, local_essent_len,
4867                                          obj->btf_modules[i].btf,
4868                                          obj->btf_modules[i].name,
4869                                          btf__get_nr_types(obj->btf_vmlinux) + 1,
4870                                          cands);
4871                 if (err)
4872                         goto err_out;
4873         }
4874
4875         return cands;
4876 err_out:
4877         bpf_core_free_cands(cands);
4878         return ERR_PTR(err);
4879 }
4880
4881 /* Check two types for compatibility for the purpose of field access
4882  * relocation. const/volatile/restrict and typedefs are skipped to ensure we
4883  * are relocating semantically compatible entities:
4884  *   - any two STRUCTs/UNIONs are compatible and can be mixed;
4885  *   - any two FWDs are compatible, if their names match (modulo flavor suffix);
4886  *   - any two PTRs are always compatible;
4887  *   - for ENUMs, names should be the same (ignoring flavor suffix) or at
4888  *     least one of enums should be anonymous;
4889  *   - for ENUMs, check sizes, names are ignored;
4890  *   - for INT, size and signedness are ignored;
4891  *   - for ARRAY, dimensionality is ignored, element types are checked for
4892  *     compatibility recursively;
4893  *   - everything else shouldn't be ever a target of relocation.
4894  * These rules are not set in stone and probably will be adjusted as we get
4895  * more experience with using BPF CO-RE relocations.
4896  */
4897 static int bpf_core_fields_are_compat(const struct btf *local_btf,
4898                                       __u32 local_id,
4899                                       const struct btf *targ_btf,
4900                                       __u32 targ_id)
4901 {
4902         const struct btf_type *local_type, *targ_type;
4903
4904 recur:
4905         local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
4906         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
4907         if (!local_type || !targ_type)
4908                 return -EINVAL;
4909
4910         if (btf_is_composite(local_type) && btf_is_composite(targ_type))
4911                 return 1;
4912         if (btf_kind(local_type) != btf_kind(targ_type))
4913                 return 0;
4914
4915         switch (btf_kind(local_type)) {
4916         case BTF_KIND_PTR:
4917                 return 1;
4918         case BTF_KIND_FWD:
4919         case BTF_KIND_ENUM: {
4920                 const char *local_name, *targ_name;
4921                 size_t local_len, targ_len;
4922
4923                 local_name = btf__name_by_offset(local_btf,
4924                                                  local_type->name_off);
4925                 targ_name = btf__name_by_offset(targ_btf, targ_type->name_off);
4926                 local_len = bpf_core_essential_name_len(local_name);
4927                 targ_len = bpf_core_essential_name_len(targ_name);
4928                 /* one of them is anonymous or both w/ same flavor-less names */
4929                 return local_len == 0 || targ_len == 0 ||
4930                        (local_len == targ_len &&
4931                         strncmp(local_name, targ_name, local_len) == 0);
4932         }
4933         case BTF_KIND_INT:
4934                 /* just reject deprecated bitfield-like integers; all other
4935                  * integers are by default compatible between each other
4936                  */
4937                 return btf_int_offset(local_type) == 0 &&
4938                        btf_int_offset(targ_type) == 0;
4939         case BTF_KIND_ARRAY:
4940                 local_id = btf_array(local_type)->type;
4941                 targ_id = btf_array(targ_type)->type;
4942                 goto recur;
4943         default:
4944                 pr_warn("unexpected kind %d relocated, local [%d], target [%d]\n",
4945                         btf_kind(local_type), local_id, targ_id);
4946                 return 0;
4947         }
4948 }
4949
4950 /*
4951  * Given single high-level named field accessor in local type, find
4952  * corresponding high-level accessor for a target type. Along the way,
4953  * maintain low-level spec for target as well. Also keep updating target
4954  * bit offset.
4955  *
4956  * Searching is performed through recursive exhaustive enumeration of all
4957  * fields of a struct/union. If there are any anonymous (embedded)
4958  * structs/unions, they are recursively searched as well. If field with
4959  * desired name is found, check compatibility between local and target types,
4960  * before returning result.
4961  *
4962  * 1 is returned, if field is found.
4963  * 0 is returned if no compatible field is found.
4964  * <0 is returned on error.
4965  */
4966 static int bpf_core_match_member(const struct btf *local_btf,
4967                                  const struct bpf_core_accessor *local_acc,
4968                                  const struct btf *targ_btf,
4969                                  __u32 targ_id,
4970                                  struct bpf_core_spec *spec,
4971                                  __u32 *next_targ_id)
4972 {
4973         const struct btf_type *local_type, *targ_type;
4974         const struct btf_member *local_member, *m;
4975         const char *local_name, *targ_name;
4976         __u32 local_id;
4977         int i, n, found;
4978
4979         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
4980         if (!targ_type)
4981                 return -EINVAL;
4982         if (!btf_is_composite(targ_type))
4983                 return 0;
4984
4985         local_id = local_acc->type_id;
4986         local_type = btf__type_by_id(local_btf, local_id);
4987         local_member = btf_members(local_type) + local_acc->idx;
4988         local_name = btf__name_by_offset(local_btf, local_member->name_off);
4989
4990         n = btf_vlen(targ_type);
4991         m = btf_members(targ_type);
4992         for (i = 0; i < n; i++, m++) {
4993                 __u32 bit_offset;
4994
4995                 bit_offset = btf_member_bit_offset(targ_type, i);
4996
4997                 /* too deep struct/union/array nesting */
4998                 if (spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
4999                         return -E2BIG;
5000
5001                 /* speculate this member will be the good one */
5002                 spec->bit_offset += bit_offset;
5003                 spec->raw_spec[spec->raw_len++] = i;
5004
5005                 targ_name = btf__name_by_offset(targ_btf, m->name_off);
5006                 if (str_is_empty(targ_name)) {
5007                         /* embedded struct/union, we need to go deeper */
5008                         found = bpf_core_match_member(local_btf, local_acc,
5009                                                       targ_btf, m->type,
5010                                                       spec, next_targ_id);
5011                         if (found) /* either found or error */
5012                                 return found;
5013                 } else if (strcmp(local_name, targ_name) == 0) {
5014                         /* matching named field */
5015                         struct bpf_core_accessor *targ_acc;
5016
5017                         targ_acc = &spec->spec[spec->len++];
5018                         targ_acc->type_id = targ_id;
5019                         targ_acc->idx = i;
5020                         targ_acc->name = targ_name;
5021
5022                         *next_targ_id = m->type;
5023                         found = bpf_core_fields_are_compat(local_btf,
5024                                                            local_member->type,
5025                                                            targ_btf, m->type);
5026                         if (!found)
5027                                 spec->len--; /* pop accessor */
5028                         return found;
5029                 }
5030                 /* member turned out not to be what we looked for */
5031                 spec->bit_offset -= bit_offset;
5032                 spec->raw_len--;
5033         }
5034
5035         return 0;
5036 }
5037
5038 /* Check local and target types for compatibility. This check is used for
5039  * type-based CO-RE relocations and follow slightly different rules than
5040  * field-based relocations. This function assumes that root types were already
5041  * checked for name match. Beyond that initial root-level name check, names
5042  * are completely ignored. Compatibility rules are as follows:
5043  *   - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but
5044  *     kind should match for local and target types (i.e., STRUCT is not
5045  *     compatible with UNION);
5046  *   - for ENUMs, the size is ignored;
5047  *   - for INT, size and signedness are ignored;
5048  *   - for ARRAY, dimensionality is ignored, element types are checked for
5049  *     compatibility recursively;
5050  *   - CONST/VOLATILE/RESTRICT modifiers are ignored;
5051  *   - TYPEDEFs/PTRs are compatible if types they pointing to are compatible;
5052  *   - FUNC_PROTOs are compatible if they have compatible signature: same
5053  *     number of input args and compatible return and argument types.
5054  * These rules are not set in stone and probably will be adjusted as we get
5055  * more experience with using BPF CO-RE relocations.
5056  */
5057 static int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id,
5058                                      const struct btf *targ_btf, __u32 targ_id)
5059 {
5060         const struct btf_type *local_type, *targ_type;
5061         int depth = 32; /* max recursion depth */
5062
5063         /* caller made sure that names match (ignoring flavor suffix) */
5064         local_type = btf__type_by_id(local_btf, local_id);
5065         targ_type = btf__type_by_id(targ_btf, targ_id);
5066         if (btf_kind(local_type) != btf_kind(targ_type))
5067                 return 0;
5068
5069 recur:
5070         depth--;
5071         if (depth < 0)
5072                 return -EINVAL;
5073
5074         local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
5075         targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
5076         if (!local_type || !targ_type)
5077                 return -EINVAL;
5078
5079         if (btf_kind(local_type) != btf_kind(targ_type))
5080                 return 0;
5081
5082         switch (btf_kind(local_type)) {
5083         case BTF_KIND_UNKN:
5084         case BTF_KIND_STRUCT:
5085         case BTF_KIND_UNION:
5086         case BTF_KIND_ENUM:
5087         case BTF_KIND_FWD:
5088                 return 1;
5089         case BTF_KIND_INT:
5090                 /* just reject deprecated bitfield-like integers; all other
5091                  * integers are by default compatible between each other
5092                  */
5093                 return btf_int_offset(local_type) == 0 && btf_int_offset(targ_type) == 0;
5094         case BTF_KIND_PTR:
5095                 local_id = local_type->type;
5096                 targ_id = targ_type->type;
5097                 goto recur;
5098         case BTF_KIND_ARRAY:
5099                 local_id = btf_array(local_type)->type;
5100                 targ_id = btf_array(targ_type)->type;
5101                 goto recur;
5102         case BTF_KIND_FUNC_PROTO: {
5103                 struct btf_param *local_p = btf_params(local_type);
5104                 struct btf_param *targ_p = btf_params(targ_type);
5105                 __u16 local_vlen = btf_vlen(local_type);
5106                 __u16 targ_vlen = btf_vlen(targ_type);
5107                 int i, err;
5108
5109                 if (local_vlen != targ_vlen)
5110                         return 0;
5111
5112                 for (i = 0; i < local_vlen; i++, local_p++, targ_p++) {
5113                         skip_mods_and_typedefs(local_btf, local_p->type, &local_id);
5114                         skip_mods_and_typedefs(targ_btf, targ_p->type, &targ_id);
5115                         err = bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id);
5116                         if (err <= 0)
5117                                 return err;
5118                 }
5119
5120                 /* tail recurse for return type check */
5121                 skip_mods_and_typedefs(local_btf, local_type->type, &local_id);
5122                 skip_mods_and_typedefs(targ_btf, targ_type->type, &targ_id);
5123                 goto recur;
5124         }
5125         default:
5126                 pr_warn("unexpected kind %s relocated, local [%d], target [%d]\n",
5127                         btf_kind_str(local_type), local_id, targ_id);
5128                 return 0;
5129         }
5130 }
5131
5132 /*
5133  * Try to match local spec to a target type and, if successful, produce full
5134  * target spec (high-level, low-level + bit offset).
5135  */
5136 static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
5137                                const struct btf *targ_btf, __u32 targ_id,
5138                                struct bpf_core_spec *targ_spec)
5139 {
5140         const struct btf_type *targ_type;
5141         const struct bpf_core_accessor *local_acc;
5142         struct bpf_core_accessor *targ_acc;
5143         int i, sz, matched;
5144
5145         memset(targ_spec, 0, sizeof(*targ_spec));
5146         targ_spec->btf = targ_btf;
5147         targ_spec->root_type_id = targ_id;
5148         targ_spec->relo_kind = local_spec->relo_kind;
5149
5150         if (core_relo_is_type_based(local_spec->relo_kind)) {
5151                 return bpf_core_types_are_compat(local_spec->btf,
5152                                                  local_spec->root_type_id,
5153                                                  targ_btf, targ_id);
5154         }
5155
5156         local_acc = &local_spec->spec[0];
5157         targ_acc = &targ_spec->spec[0];
5158
5159         if (core_relo_is_enumval_based(local_spec->relo_kind)) {
5160                 size_t local_essent_len, targ_essent_len;
5161                 const struct btf_enum *e;
5162                 const char *targ_name;
5163
5164                 /* has to resolve to an enum */
5165                 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id, &targ_id);
5166                 if (!btf_is_enum(targ_type))
5167                         return 0;
5168
5169                 local_essent_len = bpf_core_essential_name_len(local_acc->name);
5170
5171                 for (i = 0, e = btf_enum(targ_type); i < btf_vlen(targ_type); i++, e++) {
5172                         targ_name = btf__name_by_offset(targ_spec->btf, e->name_off);
5173                         targ_essent_len = bpf_core_essential_name_len(targ_name);
5174                         if (targ_essent_len != local_essent_len)
5175                                 continue;
5176                         if (strncmp(local_acc->name, targ_name, local_essent_len) == 0) {
5177                                 targ_acc->type_id = targ_id;
5178                                 targ_acc->idx = i;
5179                                 targ_acc->name = targ_name;
5180                                 targ_spec->len++;
5181                                 targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
5182                                 targ_spec->raw_len++;
5183                                 return 1;
5184                         }
5185                 }
5186                 return 0;
5187         }
5188
5189         if (!core_relo_is_field_based(local_spec->relo_kind))
5190                 return -EINVAL;
5191
5192         for (i = 0; i < local_spec->len; i++, local_acc++, targ_acc++) {
5193                 targ_type = skip_mods_and_typedefs(targ_spec->btf, targ_id,
5194                                                    &targ_id);
5195                 if (!targ_type)
5196                         return -EINVAL;
5197
5198                 if (local_acc->name) {
5199                         matched = bpf_core_match_member(local_spec->btf,
5200                                                         local_acc,
5201                                                         targ_btf, targ_id,
5202                                                         targ_spec, &targ_id);
5203                         if (matched <= 0)
5204                                 return matched;
5205                 } else {
5206                         /* for i=0, targ_id is already treated as array element
5207                          * type (because it's the original struct), for others
5208                          * we should find array element type first
5209                          */
5210                         if (i > 0) {
5211                                 const struct btf_array *a;
5212                                 bool flex;
5213
5214                                 if (!btf_is_array(targ_type))
5215                                         return 0;
5216
5217                                 a = btf_array(targ_type);
5218                                 flex = is_flex_arr(targ_btf, targ_acc - 1, a);
5219                                 if (!flex && local_acc->idx >= a->nelems)
5220                                         return 0;
5221                                 if (!skip_mods_and_typedefs(targ_btf, a->type,
5222                                                             &targ_id))
5223                                         return -EINVAL;
5224                         }
5225
5226                         /* too deep struct/union/array nesting */
5227                         if (targ_spec->raw_len == BPF_CORE_SPEC_MAX_LEN)
5228                                 return -E2BIG;
5229
5230                         targ_acc->type_id = targ_id;
5231                         targ_acc->idx = local_acc->idx;
5232                         targ_acc->name = NULL;
5233                         targ_spec->len++;
5234                         targ_spec->raw_spec[targ_spec->raw_len] = targ_acc->idx;
5235                         targ_spec->raw_len++;
5236
5237                         sz = btf__resolve_size(targ_btf, targ_id);
5238                         if (sz < 0)
5239                                 return sz;
5240                         targ_spec->bit_offset += local_acc->idx * sz * 8;
5241                 }
5242         }
5243
5244         return 1;
5245 }
5246
5247 static int bpf_core_calc_field_relo(const struct bpf_program *prog,
5248                                     const struct bpf_core_relo *relo,
5249                                     const struct bpf_core_spec *spec,
5250                                     __u32 *val, __u32 *field_sz, __u32 *type_id,
5251                                     bool *validate)
5252 {
5253         const struct bpf_core_accessor *acc;
5254         const struct btf_type *t;
5255         __u32 byte_off, byte_sz, bit_off, bit_sz, field_type_id;
5256         const struct btf_member *m;
5257         const struct btf_type *mt;
5258         bool bitfield;
5259         __s64 sz;
5260
5261         *field_sz = 0;
5262
5263         if (relo->kind == BPF_FIELD_EXISTS) {
5264                 *val = spec ? 1 : 0;
5265                 return 0;
5266         }
5267
5268         if (!spec)
5269                 return -EUCLEAN; /* request instruction poisoning */
5270
5271         acc = &spec->spec[spec->len - 1];
5272         t = btf__type_by_id(spec->btf, acc->type_id);
5273
5274         /* a[n] accessor needs special handling */
5275         if (!acc->name) {
5276                 if (relo->kind == BPF_FIELD_BYTE_OFFSET) {
5277                         *val = spec->bit_offset / 8;
5278                         /* remember field size for load/store mem size */
5279                         sz = btf__resolve_size(spec->btf, acc->type_id);
5280                         if (sz < 0)
5281                                 return -EINVAL;
5282                         *field_sz = sz;
5283                         *type_id = acc->type_id;
5284                 } else if (relo->kind == BPF_FIELD_BYTE_SIZE) {
5285                         sz = btf__resolve_size(spec->btf, acc->type_id);
5286                         if (sz < 0)
5287                                 return -EINVAL;
5288                         *val = sz;
5289                 } else {
5290                         pr_warn("prog '%s': relo %d at insn #%d can't be applied to array access\n",
5291                                 prog->name, relo->kind, relo->insn_off / 8);
5292                         return -EINVAL;
5293                 }
5294                 if (validate)
5295                         *validate = true;
5296                 return 0;
5297         }
5298
5299         m = btf_members(t) + acc->idx;
5300         mt = skip_mods_and_typedefs(spec->btf, m->type, &field_type_id);
5301         bit_off = spec->bit_offset;
5302         bit_sz = btf_member_bitfield_size(t, acc->idx);
5303
5304         bitfield = bit_sz > 0;
5305         if (bitfield) {
5306                 byte_sz = mt->size;
5307                 byte_off = bit_off / 8 / byte_sz * byte_sz;
5308                 /* figure out smallest int size necessary for bitfield load */
5309                 while (bit_off + bit_sz - byte_off * 8 > byte_sz * 8) {
5310                         if (byte_sz >= 8) {
5311                                 /* bitfield can't be read with 64-bit read */
5312                                 pr_warn("prog '%s': relo %d at insn #%d can't be satisfied for bitfield\n",
5313                                         prog->name, relo->kind, relo->insn_off / 8);
5314                                 return -E2BIG;
5315                         }
5316                         byte_sz *= 2;
5317                         byte_off = bit_off / 8 / byte_sz * byte_sz;
5318                 }
5319         } else {
5320                 sz = btf__resolve_size(spec->btf, field_type_id);
5321                 if (sz < 0)
5322                         return -EINVAL;
5323                 byte_sz = sz;
5324                 byte_off = spec->bit_offset / 8;
5325                 bit_sz = byte_sz * 8;
5326         }
5327
5328         /* for bitfields, all the relocatable aspects are ambiguous and we
5329          * might disagree with compiler, so turn off validation of expected
5330          * value, except for signedness
5331          */
5332         if (validate)
5333                 *validate = !bitfield;
5334
5335         switch (relo->kind) {
5336         case BPF_FIELD_BYTE_OFFSET:
5337                 *val = byte_off;
5338                 if (!bitfield) {
5339                         *field_sz = byte_sz;
5340                         *type_id = field_type_id;
5341                 }
5342                 break;
5343         case BPF_FIELD_BYTE_SIZE:
5344                 *val = byte_sz;
5345                 break;
5346         case BPF_FIELD_SIGNED:
5347                 /* enums will be assumed unsigned */
5348                 *val = btf_is_enum(mt) ||
5349                        (btf_int_encoding(mt) & BTF_INT_SIGNED);
5350                 if (validate)
5351                         *validate = true; /* signedness is never ambiguous */
5352                 break;
5353         case BPF_FIELD_LSHIFT_U64:
5354 #if __BYTE_ORDER == __LITTLE_ENDIAN
5355                 *val = 64 - (bit_off + bit_sz - byte_off  * 8);
5356 #else
5357                 *val = (8 - byte_sz) * 8 + (bit_off - byte_off * 8);
5358 #endif
5359                 break;
5360         case BPF_FIELD_RSHIFT_U64:
5361                 *val = 64 - bit_sz;
5362                 if (validate)
5363                         *validate = true; /* right shift is never ambiguous */
5364                 break;
5365         case BPF_FIELD_EXISTS:
5366         default:
5367                 return -EOPNOTSUPP;
5368         }
5369
5370         return 0;
5371 }
5372
5373 static int bpf_core_calc_type_relo(const struct bpf_core_relo *relo,
5374                                    const struct bpf_core_spec *spec,
5375                                    __u32 *val)
5376 {
5377         __s64 sz;
5378
5379         /* type-based relos return zero when target type is not found */
5380         if (!spec) {
5381                 *val = 0;
5382                 return 0;
5383         }
5384
5385         switch (relo->kind) {
5386         case BPF_TYPE_ID_TARGET:
5387                 *val = spec->root_type_id;
5388                 break;
5389         case BPF_TYPE_EXISTS:
5390                 *val = 1;
5391                 break;
5392         case BPF_TYPE_SIZE:
5393                 sz = btf__resolve_size(spec->btf, spec->root_type_id);
5394                 if (sz < 0)
5395                         return -EINVAL;
5396                 *val = sz;
5397                 break;
5398         case BPF_TYPE_ID_LOCAL:
5399         /* BPF_TYPE_ID_LOCAL is handled specially and shouldn't get here */
5400         default:
5401                 return -EOPNOTSUPP;
5402         }
5403
5404         return 0;
5405 }
5406
5407 static int bpf_core_calc_enumval_relo(const struct bpf_core_relo *relo,
5408                                       const struct bpf_core_spec *spec,
5409                                       __u32 *val)
5410 {
5411         const struct btf_type *t;
5412         const struct btf_enum *e;
5413
5414         switch (relo->kind) {
5415         case BPF_ENUMVAL_EXISTS:
5416                 *val = spec ? 1 : 0;
5417                 break;
5418         case BPF_ENUMVAL_VALUE:
5419                 if (!spec)
5420                         return -EUCLEAN; /* request instruction poisoning */
5421                 t = btf__type_by_id(spec->btf, spec->spec[0].type_id);
5422                 e = btf_enum(t) + spec->spec[0].idx;
5423                 *val = e->val;
5424                 break;
5425         default:
5426                 return -EOPNOTSUPP;
5427         }
5428
5429         return 0;
5430 }
5431
5432 struct bpf_core_relo_res
5433 {
5434         /* expected value in the instruction, unless validate == false */
5435         __u32 orig_val;
5436         /* new value that needs to be patched up to */
5437         __u32 new_val;
5438         /* relocation unsuccessful, poison instruction, but don't fail load */
5439         bool poison;
5440         /* some relocations can't be validated against orig_val */
5441         bool validate;
5442         /* for field byte offset relocations or the forms:
5443          *     *(T *)(rX + <off>) = rY
5444          *     rX = *(T *)(rY + <off>),
5445          * we remember original and resolved field size to adjust direct
5446          * memory loads of pointers and integers; this is necessary for 32-bit
5447          * host kernel architectures, but also allows to automatically
5448          * relocate fields that were resized from, e.g., u32 to u64, etc.
5449          */
5450         bool fail_memsz_adjust;
5451         __u32 orig_sz;
5452         __u32 orig_type_id;
5453         __u32 new_sz;
5454         __u32 new_type_id;
5455 };
5456
5457 /* Calculate original and target relocation values, given local and target
5458  * specs and relocation kind. These values are calculated for each candidate.
5459  * If there are multiple candidates, resulting values should all be consistent
5460  * with each other. Otherwise, libbpf will refuse to proceed due to ambiguity.
5461  * If instruction has to be poisoned, *poison will be set to true.
5462  */
5463 static int bpf_core_calc_relo(const struct bpf_program *prog,
5464                               const struct bpf_core_relo *relo,
5465                               int relo_idx,
5466                               const struct bpf_core_spec *local_spec,
5467                               const struct bpf_core_spec *targ_spec,
5468                               struct bpf_core_relo_res *res)
5469 {
5470         int err = -EOPNOTSUPP;
5471
5472         res->orig_val = 0;
5473         res->new_val = 0;
5474         res->poison = false;
5475         res->validate = true;
5476         res->fail_memsz_adjust = false;
5477         res->orig_sz = res->new_sz = 0;
5478         res->orig_type_id = res->new_type_id = 0;
5479
5480         if (core_relo_is_field_based(relo->kind)) {
5481                 err = bpf_core_calc_field_relo(prog, relo, local_spec,
5482                                                &res->orig_val, &res->orig_sz,
5483                                                &res->orig_type_id, &res->validate);
5484                 err = err ?: bpf_core_calc_field_relo(prog, relo, targ_spec,
5485                                                       &res->new_val, &res->new_sz,
5486                                                       &res->new_type_id, NULL);
5487                 if (err)
5488                         goto done;
5489                 /* Validate if it's safe to adjust load/store memory size.
5490                  * Adjustments are performed only if original and new memory
5491                  * sizes differ.
5492                  */
5493                 res->fail_memsz_adjust = false;
5494                 if (res->orig_sz != res->new_sz) {
5495                         const struct btf_type *orig_t, *new_t;
5496
5497                         orig_t = btf__type_by_id(local_spec->btf, res->orig_type_id);
5498                         new_t = btf__type_by_id(targ_spec->btf, res->new_type_id);
5499
5500                         /* There are two use cases in which it's safe to
5501                          * adjust load/store's mem size:
5502                          *   - reading a 32-bit kernel pointer, while on BPF
5503                          *   size pointers are always 64-bit; in this case
5504                          *   it's safe to "downsize" instruction size due to
5505                          *   pointer being treated as unsigned integer with
5506                          *   zero-extended upper 32-bits;
5507                          *   - reading unsigned integers, again due to
5508                          *   zero-extension is preserving the value correctly.
5509                          *
5510                          * In all other cases it's incorrect to attempt to
5511                          * load/store field because read value will be
5512                          * incorrect, so we poison relocated instruction.
5513                          */
5514                         if (btf_is_ptr(orig_t) && btf_is_ptr(new_t))
5515                                 goto done;
5516                         if (btf_is_int(orig_t) && btf_is_int(new_t) &&
5517                             btf_int_encoding(orig_t) != BTF_INT_SIGNED &&
5518                             btf_int_encoding(new_t) != BTF_INT_SIGNED)
5519                                 goto done;
5520
5521                         /* mark as invalid mem size adjustment, but this will
5522                          * only be checked for LDX/STX/ST insns
5523                          */
5524                         res->fail_memsz_adjust = true;
5525                 }
5526         } else if (core_relo_is_type_based(relo->kind)) {
5527                 err = bpf_core_calc_type_relo(relo, local_spec, &res->orig_val);
5528                 err = err ?: bpf_core_calc_type_relo(relo, targ_spec, &res->new_val);
5529         } else if (core_relo_is_enumval_based(relo->kind)) {
5530                 err = bpf_core_calc_enumval_relo(relo, local_spec, &res->orig_val);
5531                 err = err ?: bpf_core_calc_enumval_relo(relo, targ_spec, &res->new_val);
5532         }
5533
5534 done:
5535         if (err == -EUCLEAN) {
5536                 /* EUCLEAN is used to signal instruction poisoning request */
5537                 res->poison = true;
5538                 err = 0;
5539         } else if (err == -EOPNOTSUPP) {
5540                 /* EOPNOTSUPP means unknown/unsupported relocation */
5541                 pr_warn("prog '%s': relo #%d: unrecognized CO-RE relocation %s (%d) at insn #%d\n",
5542                         prog->name, relo_idx, core_relo_kind_str(relo->kind),
5543                         relo->kind, relo->insn_off / 8);
5544         }
5545
5546         return err;
5547 }
5548
5549 /*
5550  * Turn instruction for which CO_RE relocation failed into invalid one with
5551  * distinct signature.
5552  */
5553 static void bpf_core_poison_insn(struct bpf_program *prog, int relo_idx,
5554                                  int insn_idx, struct bpf_insn *insn)
5555 {
5556         pr_debug("prog '%s': relo #%d: substituting insn #%d w/ invalid insn\n",
5557                  prog->name, relo_idx, insn_idx);
5558         insn->code = BPF_JMP | BPF_CALL;
5559         insn->dst_reg = 0;
5560         insn->src_reg = 0;
5561         insn->off = 0;
5562         /* if this instruction is reachable (not a dead code),
5563          * verifier will complain with the following message:
5564          * invalid func unknown#195896080
5565          */
5566         insn->imm = 195896080; /* => 0xbad2310 => "bad relo" */
5567 }
5568
5569 static bool is_ldimm64(struct bpf_insn *insn)
5570 {
5571         return insn->code == (BPF_LD | BPF_IMM | BPF_DW);
5572 }
5573
5574 static int insn_bpf_size_to_bytes(struct bpf_insn *insn)
5575 {
5576         switch (BPF_SIZE(insn->code)) {
5577         case BPF_DW: return 8;
5578         case BPF_W: return 4;
5579         case BPF_H: return 2;
5580         case BPF_B: return 1;
5581         default: return -1;
5582         }
5583 }
5584
5585 static int insn_bytes_to_bpf_size(__u32 sz)
5586 {
5587         switch (sz) {
5588         case 8: return BPF_DW;
5589         case 4: return BPF_W;
5590         case 2: return BPF_H;
5591         case 1: return BPF_B;
5592         default: return -1;
5593         }
5594 }
5595
5596 /*
5597  * Patch relocatable BPF instruction.
5598  *
5599  * Patched value is determined by relocation kind and target specification.
5600  * For existence relocations target spec will be NULL if field/type is not found.
5601  * Expected insn->imm value is determined using relocation kind and local
5602  * spec, and is checked before patching instruction. If actual insn->imm value
5603  * is wrong, bail out with error.
5604  *
5605  * Currently supported classes of BPF instruction are:
5606  * 1. rX = <imm> (assignment with immediate operand);
5607  * 2. rX += <imm> (arithmetic operations with immediate operand);
5608  * 3. rX = <imm64> (load with 64-bit immediate value);
5609  * 4. rX = *(T *)(rY + <off>), where T is one of {u8, u16, u32, u64};
5610  * 5. *(T *)(rX + <off>) = rY, where T is one of {u8, u16, u32, u64};
5611  * 6. *(T *)(rX + <off>) = <imm>, where T is one of {u8, u16, u32, u64}.
5612  */
5613 static int bpf_core_patch_insn(struct bpf_program *prog,
5614                                const struct bpf_core_relo *relo,
5615                                int relo_idx,
5616                                const struct bpf_core_relo_res *res)
5617 {
5618         __u32 orig_val, new_val;
5619         struct bpf_insn *insn;
5620         int insn_idx;
5621         __u8 class;
5622
5623         if (relo->insn_off % BPF_INSN_SZ)
5624                 return -EINVAL;
5625         insn_idx = relo->insn_off / BPF_INSN_SZ;
5626         /* adjust insn_idx from section frame of reference to the local
5627          * program's frame of reference; (sub-)program code is not yet
5628          * relocated, so it's enough to just subtract in-section offset
5629          */
5630         insn_idx = insn_idx - prog->sec_insn_off;
5631         insn = &prog->insns[insn_idx];
5632         class = BPF_CLASS(insn->code);
5633
5634         if (res->poison) {
5635 poison:
5636                 /* poison second part of ldimm64 to avoid confusing error from
5637                  * verifier about "unknown opcode 00"
5638                  */
5639                 if (is_ldimm64(insn))
5640                         bpf_core_poison_insn(prog, relo_idx, insn_idx + 1, insn + 1);
5641                 bpf_core_poison_insn(prog, relo_idx, insn_idx, insn);
5642                 return 0;
5643         }
5644
5645         orig_val = res->orig_val;
5646         new_val = res->new_val;
5647
5648         switch (class) {
5649         case BPF_ALU:
5650         case BPF_ALU64:
5651                 if (BPF_SRC(insn->code) != BPF_K)
5652                         return -EINVAL;
5653                 if (res->validate && insn->imm != orig_val) {
5654                         pr_warn("prog '%s': relo #%d: unexpected insn #%d (ALU/ALU64) value: got %u, exp %u -> %u\n",
5655                                 prog->name, relo_idx,
5656                                 insn_idx, insn->imm, orig_val, new_val);
5657                         return -EINVAL;
5658                 }
5659                 orig_val = insn->imm;
5660                 insn->imm = new_val;
5661                 pr_debug("prog '%s': relo #%d: patched insn #%d (ALU/ALU64) imm %u -> %u\n",
5662                          prog->name, relo_idx, insn_idx,
5663                          orig_val, new_val);
5664                 break;
5665         case BPF_LDX:
5666         case BPF_ST:
5667         case BPF_STX:
5668                 if (res->validate && insn->off != orig_val) {
5669                         pr_warn("prog '%s': relo #%d: unexpected insn #%d (LDX/ST/STX) value: got %u, exp %u -> %u\n",
5670                                 prog->name, relo_idx, insn_idx, insn->off, orig_val, new_val);
5671                         return -EINVAL;
5672                 }
5673                 if (new_val > SHRT_MAX) {
5674                         pr_warn("prog '%s': relo #%d: insn #%d (LDX/ST/STX) value too big: %u\n",
5675                                 prog->name, relo_idx, insn_idx, new_val);
5676                         return -ERANGE;
5677                 }
5678                 if (res->fail_memsz_adjust) {
5679                         pr_warn("prog '%s': relo #%d: insn #%d (LDX/ST/STX) accesses field incorrectly. "
5680                                 "Make sure you are accessing pointers, unsigned integers, or fields of matching type and size.\n",
5681                                 prog->name, relo_idx, insn_idx);
5682                         goto poison;
5683                 }
5684
5685                 orig_val = insn->off;
5686                 insn->off = new_val;
5687                 pr_debug("prog '%s': relo #%d: patched insn #%d (LDX/ST/STX) off %u -> %u\n",
5688                          prog->name, relo_idx, insn_idx, orig_val, new_val);
5689
5690                 if (res->new_sz != res->orig_sz) {
5691                         int insn_bytes_sz, insn_bpf_sz;
5692
5693                         insn_bytes_sz = insn_bpf_size_to_bytes(insn);
5694                         if (insn_bytes_sz != res->orig_sz) {
5695                                 pr_warn("prog '%s': relo #%d: insn #%d (LDX/ST/STX) unexpected mem size: got %d, exp %u\n",
5696                                         prog->name, relo_idx, insn_idx, insn_bytes_sz, res->orig_sz);
5697                                 return -EINVAL;
5698                         }
5699
5700                         insn_bpf_sz = insn_bytes_to_bpf_size(res->new_sz);
5701                         if (insn_bpf_sz < 0) {
5702                                 pr_warn("prog '%s': relo #%d: insn #%d (LDX/ST/STX) invalid new mem size: %u\n",
5703                                         prog->name, relo_idx, insn_idx, res->new_sz);
5704                                 return -EINVAL;
5705                         }
5706
5707                         insn->code = BPF_MODE(insn->code) | insn_bpf_sz | BPF_CLASS(insn->code);
5708                         pr_debug("prog '%s': relo #%d: patched insn #%d (LDX/ST/STX) mem_sz %u -> %u\n",
5709                                  prog->name, relo_idx, insn_idx, res->orig_sz, res->new_sz);
5710                 }
5711                 break;
5712         case BPF_LD: {
5713                 __u64 imm;
5714
5715                 if (!is_ldimm64(insn) ||
5716                     insn[0].src_reg != 0 || insn[0].off != 0 ||
5717                     insn_idx + 1 >= prog->insns_cnt ||
5718                     insn[1].code != 0 || insn[1].dst_reg != 0 ||
5719                     insn[1].src_reg != 0 || insn[1].off != 0) {
5720                         pr_warn("prog '%s': relo #%d: insn #%d (LDIMM64) has unexpected form\n",
5721                                 prog->name, relo_idx, insn_idx);
5722                         return -EINVAL;
5723                 }
5724
5725                 imm = insn[0].imm + ((__u64)insn[1].imm << 32);
5726                 if (res->validate && imm != orig_val) {
5727                         pr_warn("prog '%s': relo #%d: unexpected insn #%d (LDIMM64) value: got %llu, exp %u -> %u\n",
5728                                 prog->name, relo_idx,
5729                                 insn_idx, (unsigned long long)imm,
5730                                 orig_val, new_val);
5731                         return -EINVAL;
5732                 }
5733
5734                 insn[0].imm = new_val;
5735                 insn[1].imm = 0; /* currently only 32-bit values are supported */
5736                 pr_debug("prog '%s': relo #%d: patched insn #%d (LDIMM64) imm64 %llu -> %u\n",
5737                          prog->name, relo_idx, insn_idx,
5738                          (unsigned long long)imm, new_val);
5739                 break;
5740         }
5741         default:
5742                 pr_warn("prog '%s': relo #%d: trying to relocate unrecognized insn #%d, code:0x%x, src:0x%x, dst:0x%x, off:0x%x, imm:0x%x\n",
5743                         prog->name, relo_idx, insn_idx, insn->code,
5744                         insn->src_reg, insn->dst_reg, insn->off, insn->imm);
5745                 return -EINVAL;
5746         }
5747
5748         return 0;
5749 }
5750
5751 /* Output spec definition in the format:
5752  * [<type-id>] (<type-name>) + <raw-spec> => <offset>@<spec>,
5753  * where <spec> is a C-syntax view of recorded field access, e.g.: x.a[3].b
5754  */
5755 static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
5756 {
5757         const struct btf_type *t;
5758         const struct btf_enum *e;
5759         const char *s;
5760         __u32 type_id;
5761         int i;
5762
5763         type_id = spec->root_type_id;
5764         t = btf__type_by_id(spec->btf, type_id);
5765         s = btf__name_by_offset(spec->btf, t->name_off);
5766
5767         libbpf_print(level, "[%u] %s %s", type_id, btf_kind_str(t), str_is_empty(s) ? "<anon>" : s);
5768
5769         if (core_relo_is_type_based(spec->relo_kind))
5770                 return;
5771
5772         if (core_relo_is_enumval_based(spec->relo_kind)) {
5773                 t = skip_mods_and_typedefs(spec->btf, type_id, NULL);
5774                 e = btf_enum(t) + spec->raw_spec[0];
5775                 s = btf__name_by_offset(spec->btf, e->name_off);
5776
5777                 libbpf_print(level, "::%s = %u", s, e->val);
5778                 return;
5779         }
5780
5781         if (core_relo_is_field_based(spec->relo_kind)) {
5782                 for (i = 0; i < spec->len; i++) {
5783                         if (spec->spec[i].name)
5784                                 libbpf_print(level, ".%s", spec->spec[i].name);
5785                         else if (i > 0 || spec->spec[i].idx > 0)
5786                                 libbpf_print(level, "[%u]", spec->spec[i].idx);
5787                 }
5788
5789                 libbpf_print(level, " (");
5790                 for (i = 0; i < spec->raw_len; i++)
5791                         libbpf_print(level, "%s%d", i == 0 ? "" : ":", spec->raw_spec[i]);
5792
5793                 if (spec->bit_offset % 8)
5794                         libbpf_print(level, " @ offset %u.%u)",
5795                                      spec->bit_offset / 8, spec->bit_offset % 8);
5796                 else
5797                         libbpf_print(level, " @ offset %u)", spec->bit_offset / 8);
5798                 return;
5799         }
5800 }
5801
5802 static size_t bpf_core_hash_fn(const void *key, void *ctx)
5803 {
5804         return (size_t)key;
5805 }
5806
5807 static bool bpf_core_equal_fn(const void *k1, const void *k2, void *ctx)
5808 {
5809         return k1 == k2;
5810 }
5811
5812 static void *u32_as_hash_key(__u32 x)
5813 {
5814         return (void *)(uintptr_t)x;
5815 }
5816
5817 /*
5818  * CO-RE relocate single instruction.
5819  *
5820  * The outline and important points of the algorithm:
5821  * 1. For given local type, find corresponding candidate target types.
5822  *    Candidate type is a type with the same "essential" name, ignoring
5823  *    everything after last triple underscore (___). E.g., `sample`,
5824  *    `sample___flavor_one`, `sample___flavor_another_one`, are all candidates
5825  *    for each other. Names with triple underscore are referred to as
5826  *    "flavors" and are useful, among other things, to allow to
5827  *    specify/support incompatible variations of the same kernel struct, which
5828  *    might differ between different kernel versions and/or build
5829  *    configurations.
5830  *
5831  *    N.B. Struct "flavors" could be generated by bpftool's BTF-to-C
5832  *    converter, when deduplicated BTF of a kernel still contains more than
5833  *    one different types with the same name. In that case, ___2, ___3, etc
5834  *    are appended starting from second name conflict. But start flavors are
5835  *    also useful to be defined "locally", in BPF program, to extract same
5836  *    data from incompatible changes between different kernel
5837  *    versions/configurations. For instance, to handle field renames between
5838  *    kernel versions, one can use two flavors of the struct name with the
5839  *    same common name and use conditional relocations to extract that field,
5840  *    depending on target kernel version.
5841  * 2. For each candidate type, try to match local specification to this
5842  *    candidate target type. Matching involves finding corresponding
5843  *    high-level spec accessors, meaning that all named fields should match,
5844  *    as well as all array accesses should be within the actual bounds. Also,
5845  *    types should be compatible (see bpf_core_fields_are_compat for details).
5846  * 3. It is supported and expected that there might be multiple flavors
5847  *    matching the spec. As long as all the specs resolve to the same set of
5848  *    offsets across all candidates, there is no error. If there is any
5849  *    ambiguity, CO-RE relocation will fail. This is necessary to accomodate
5850  *    imprefection of BTF deduplication, which can cause slight duplication of
5851  *    the same BTF type, if some directly or indirectly referenced (by
5852  *    pointer) type gets resolved to different actual types in different
5853  *    object files. If such situation occurs, deduplicated BTF will end up
5854  *    with two (or more) structurally identical types, which differ only in
5855  *    types they refer to through pointer. This should be OK in most cases and
5856  *    is not an error.
5857  * 4. Candidate types search is performed by linearly scanning through all
5858  *    types in target BTF. It is anticipated that this is overall more
5859  *    efficient memory-wise and not significantly worse (if not better)
5860  *    CPU-wise compared to prebuilding a map from all local type names to
5861  *    a list of candidate type names. It's also sped up by caching resolved
5862  *    list of matching candidates per each local "root" type ID, that has at
5863  *    least one bpf_core_relo associated with it. This list is shared
5864  *    between multiple relocations for the same type ID and is updated as some
5865  *    of the candidates are pruned due to structural incompatibility.
5866  */
5867 static int bpf_core_apply_relo(struct bpf_program *prog,
5868                                const struct bpf_core_relo *relo,
5869                                int relo_idx,
5870                                const struct btf *local_btf,
5871                                struct hashmap *cand_cache)
5872 {
5873         struct bpf_core_spec local_spec, cand_spec, targ_spec = {};
5874         const void *type_key = u32_as_hash_key(relo->type_id);
5875         struct bpf_core_relo_res cand_res, targ_res;
5876         const struct btf_type *local_type;
5877         const char *local_name;
5878         struct core_cand_list *cands = NULL;
5879         __u32 local_id;
5880         const char *spec_str;
5881         int i, j, err;
5882
5883         local_id = relo->type_id;
5884         local_type = btf__type_by_id(local_btf, local_id);
5885         if (!local_type)
5886                 return -EINVAL;
5887
5888         local_name = btf__name_by_offset(local_btf, local_type->name_off);
5889         if (!local_name)
5890                 return -EINVAL;
5891
5892         spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
5893         if (str_is_empty(spec_str))
5894                 return -EINVAL;
5895
5896         err = bpf_core_parse_spec(local_btf, local_id, spec_str, relo->kind, &local_spec);
5897         if (err) {
5898                 pr_warn("prog '%s': relo #%d: parsing [%d] %s %s + %s failed: %d\n",
5899                         prog->name, relo_idx, local_id, btf_kind_str(local_type),
5900                         str_is_empty(local_name) ? "<anon>" : local_name,
5901                         spec_str, err);
5902                 return -EINVAL;
5903         }
5904
5905         pr_debug("prog '%s': relo #%d: kind <%s> (%d), spec is ", prog->name,
5906                  relo_idx, core_relo_kind_str(relo->kind), relo->kind);
5907         bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
5908         libbpf_print(LIBBPF_DEBUG, "\n");
5909
5910         /* TYPE_ID_LOCAL relo is special and doesn't need candidate search */
5911         if (relo->kind == BPF_TYPE_ID_LOCAL) {
5912                 targ_res.validate = true;
5913                 targ_res.poison = false;
5914                 targ_res.orig_val = local_spec.root_type_id;
5915                 targ_res.new_val = local_spec.root_type_id;
5916                 goto patch_insn;
5917         }
5918
5919         /* libbpf doesn't support candidate search for anonymous types */
5920         if (str_is_empty(spec_str)) {
5921                 pr_warn("prog '%s': relo #%d: <%s> (%d) relocation doesn't support anonymous types\n",
5922                         prog->name, relo_idx, core_relo_kind_str(relo->kind), relo->kind);
5923                 return -EOPNOTSUPP;
5924         }
5925
5926         if (!hashmap__find(cand_cache, type_key, (void **)&cands)) {
5927                 cands = bpf_core_find_cands(prog->obj, local_btf, local_id);
5928                 if (IS_ERR(cands)) {
5929                         pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n",
5930                                 prog->name, relo_idx, local_id, btf_kind_str(local_type),
5931                                 local_name, PTR_ERR(cands));
5932                         return PTR_ERR(cands);
5933                 }
5934                 err = hashmap__set(cand_cache, type_key, cands, NULL, NULL);
5935                 if (err) {
5936                         bpf_core_free_cands(cands);
5937                         return err;
5938                 }
5939         }
5940
5941         for (i = 0, j = 0; i < cands->len; i++) {
5942                 err = bpf_core_spec_match(&local_spec, cands->cands[i].btf,
5943                                           cands->cands[i].id, &cand_spec);
5944                 if (err < 0) {
5945                         pr_warn("prog '%s': relo #%d: error matching candidate #%d ",
5946                                 prog->name, relo_idx, i);
5947                         bpf_core_dump_spec(LIBBPF_WARN, &cand_spec);
5948                         libbpf_print(LIBBPF_WARN, ": %d\n", err);
5949                         return err;
5950                 }
5951
5952                 pr_debug("prog '%s': relo #%d: %s candidate #%d ", prog->name,
5953                          relo_idx, err == 0 ? "non-matching" : "matching", i);
5954                 bpf_core_dump_spec(LIBBPF_DEBUG, &cand_spec);
5955                 libbpf_print(LIBBPF_DEBUG, "\n");
5956
5957                 if (err == 0)
5958                         continue;
5959
5960                 err = bpf_core_calc_relo(prog, relo, relo_idx, &local_spec, &cand_spec, &cand_res);
5961                 if (err)
5962                         return err;
5963
5964                 if (j == 0) {
5965                         targ_res = cand_res;
5966                         targ_spec = cand_spec;
5967                 } else if (cand_spec.bit_offset != targ_spec.bit_offset) {
5968                         /* if there are many field relo candidates, they
5969                          * should all resolve to the same bit offset
5970                          */
5971                         pr_warn("prog '%s': relo #%d: field offset ambiguity: %u != %u\n",
5972                                 prog->name, relo_idx, cand_spec.bit_offset,
5973                                 targ_spec.bit_offset);
5974                         return -EINVAL;
5975                 } else if (cand_res.poison != targ_res.poison || cand_res.new_val != targ_res.new_val) {
5976                         /* all candidates should result in the same relocation
5977                          * decision and value, otherwise it's dangerous to
5978                          * proceed due to ambiguity
5979                          */
5980                         pr_warn("prog '%s': relo #%d: relocation decision ambiguity: %s %u != %s %u\n",
5981                                 prog->name, relo_idx,
5982                                 cand_res.poison ? "failure" : "success", cand_res.new_val,
5983                                 targ_res.poison ? "failure" : "success", targ_res.new_val);
5984                         return -EINVAL;
5985                 }
5986
5987                 cands->cands[j++] = cands->cands[i];
5988         }
5989
5990         /*
5991          * For BPF_FIELD_EXISTS relo or when used BPF program has field
5992          * existence checks or kernel version/config checks, it's expected
5993          * that we might not find any candidates. In this case, if field
5994          * wasn't found in any candidate, the list of candidates shouldn't
5995          * change at all, we'll just handle relocating appropriately,
5996          * depending on relo's kind.
5997          */
5998         if (j > 0)
5999                 cands->len = j;
6000
6001         /*
6002          * If no candidates were found, it might be both a programmer error,
6003          * as well as expected case, depending whether instruction w/
6004          * relocation is guarded in some way that makes it unreachable (dead
6005          * code) if relocation can't be resolved. This is handled in
6006          * bpf_core_patch_insn() uniformly by replacing that instruction with
6007          * BPF helper call insn (using invalid helper ID). If that instruction
6008          * is indeed unreachable, then it will be ignored and eliminated by
6009          * verifier. If it was an error, then verifier will complain and point
6010          * to a specific instruction number in its log.
6011          */
6012         if (j == 0) {
6013                 pr_debug("prog '%s': relo #%d: no matching targets found\n",
6014                          prog->name, relo_idx);
6015
6016                 /* calculate single target relo result explicitly */
6017                 err = bpf_core_calc_relo(prog, relo, relo_idx, &local_spec, NULL, &targ_res);
6018                 if (err)
6019                         return err;
6020         }
6021
6022 patch_insn:
6023         /* bpf_core_patch_insn() should know how to handle missing targ_spec */
6024         err = bpf_core_patch_insn(prog, relo, relo_idx, &targ_res);
6025         if (err) {
6026                 pr_warn("prog '%s': relo #%d: failed to patch insn at offset %d: %d\n",
6027                         prog->name, relo_idx, relo->insn_off, err);
6028                 return -EINVAL;
6029         }
6030
6031         return 0;
6032 }
6033
6034 static int
6035 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
6036 {
6037         const struct btf_ext_info_sec *sec;
6038         const struct bpf_core_relo *rec;
6039         const struct btf_ext_info *seg;
6040         struct hashmap_entry *entry;
6041         struct hashmap *cand_cache = NULL;
6042         struct bpf_program *prog;
6043         const char *sec_name;
6044         int i, err = 0, insn_idx, sec_idx;
6045
6046         if (obj->btf_ext->core_relo_info.len == 0)
6047                 return 0;
6048
6049         if (targ_btf_path) {
6050                 obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL);
6051                 if (IS_ERR_OR_NULL(obj->btf_vmlinux_override)) {
6052                         err = PTR_ERR(obj->btf_vmlinux_override);
6053                         pr_warn("failed to parse target BTF: %d\n", err);
6054                         return err;
6055                 }
6056         }
6057
6058         cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
6059         if (IS_ERR(cand_cache)) {
6060                 err = PTR_ERR(cand_cache);
6061                 goto out;
6062         }
6063
6064         seg = &obj->btf_ext->core_relo_info;
6065         for_each_btf_ext_sec(seg, sec) {
6066                 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
6067                 if (str_is_empty(sec_name)) {
6068                         err = -EINVAL;
6069                         goto out;
6070                 }
6071                 /* bpf_object's ELF is gone by now so it's not easy to find
6072                  * section index by section name, but we can find *any*
6073                  * bpf_program within desired section name and use it's
6074                  * prog->sec_idx to do a proper search by section index and
6075                  * instruction offset
6076                  */
6077                 prog = NULL;
6078                 for (i = 0; i < obj->nr_programs; i++) {
6079                         prog = &obj->programs[i];
6080                         if (strcmp(prog->sec_name, sec_name) == 0)
6081                                 break;
6082                 }
6083                 if (!prog) {
6084                         pr_warn("sec '%s': failed to find a BPF program\n", sec_name);
6085                         return -ENOENT;
6086                 }
6087                 sec_idx = prog->sec_idx;
6088
6089                 pr_debug("sec '%s': found %d CO-RE relocations\n",
6090                          sec_name, sec->num_info);
6091
6092                 for_each_btf_ext_rec(seg, sec, i, rec) {
6093                         insn_idx = rec->insn_off / BPF_INSN_SZ;
6094                         prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
6095                         if (!prog) {
6096                                 pr_warn("sec '%s': failed to find program at insn #%d for CO-RE offset relocation #%d\n",
6097                                         sec_name, insn_idx, i);
6098                                 err = -EINVAL;
6099                                 goto out;
6100                         }
6101                         /* no need to apply CO-RE relocation if the program is
6102                          * not going to be loaded
6103                          */
6104                         if (!prog->load)
6105                                 continue;
6106
6107                         err = bpf_core_apply_relo(prog, rec, i, obj->btf, cand_cache);
6108                         if (err) {
6109                                 pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
6110                                         prog->name, i, err);
6111                                 goto out;
6112                         }
6113                 }
6114         }
6115
6116 out:
6117         /* obj->btf_vmlinux and module BTFs are freed after object load */
6118         btf__free(obj->btf_vmlinux_override);
6119         obj->btf_vmlinux_override = NULL;
6120
6121         if (!IS_ERR_OR_NULL(cand_cache)) {
6122                 hashmap__for_each_entry(cand_cache, entry, i) {
6123                         bpf_core_free_cands(entry->value);
6124                 }
6125                 hashmap__free(cand_cache);
6126         }
6127         return err;
6128 }
6129
6130 /* Relocate data references within program code:
6131  *  - map references;
6132  *  - global variable references;
6133  *  - extern references.
6134  */
6135 static int
6136 bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
6137 {
6138         int i;
6139
6140         for (i = 0; i < prog->nr_reloc; i++) {
6141                 struct reloc_desc *relo = &prog->reloc_desc[i];
6142                 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
6143                 struct extern_desc *ext;
6144
6145                 switch (relo->type) {
6146                 case RELO_LD64:
6147                         insn[0].src_reg = BPF_PSEUDO_MAP_FD;
6148                         insn[0].imm = obj->maps[relo->map_idx].fd;
6149                         relo->processed = true;
6150                         break;
6151                 case RELO_DATA:
6152                         insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
6153                         insn[1].imm = insn[0].imm + relo->sym_off;
6154                         insn[0].imm = obj->maps[relo->map_idx].fd;
6155                         relo->processed = true;
6156                         break;
6157                 case RELO_EXTERN:
6158                         ext = &obj->externs[relo->sym_off];
6159                         if (ext->type == EXT_KCFG) {
6160                                 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
6161                                 insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
6162                                 insn[1].imm = ext->kcfg.data_off;
6163                         } else /* EXT_KSYM */ {
6164                                 if (ext->ksym.type_id) { /* typed ksyms */
6165                                         insn[0].src_reg = BPF_PSEUDO_BTF_ID;
6166                                         insn[0].imm = ext->ksym.kernel_btf_id;
6167                                         insn[1].imm = ext->ksym.kernel_btf_obj_fd;
6168                                 } else { /* typeless ksyms */
6169                                         insn[0].imm = (__u32)ext->ksym.addr;
6170                                         insn[1].imm = ext->ksym.addr >> 32;
6171                                 }
6172                         }
6173                         relo->processed = true;
6174                         break;
6175                 case RELO_CALL:
6176                         /* will be handled as a follow up pass */
6177                         break;
6178                 default:
6179                         pr_warn("prog '%s': relo #%d: bad relo type %d\n",
6180                                 prog->name, i, relo->type);
6181                         return -EINVAL;
6182                 }
6183         }
6184
6185         return 0;
6186 }
6187
6188 static int adjust_prog_btf_ext_info(const struct bpf_object *obj,
6189                                     const struct bpf_program *prog,
6190                                     const struct btf_ext_info *ext_info,
6191                                     void **prog_info, __u32 *prog_rec_cnt,
6192                                     __u32 *prog_rec_sz)
6193 {
6194         void *copy_start = NULL, *copy_end = NULL;
6195         void *rec, *rec_end, *new_prog_info;
6196         const struct btf_ext_info_sec *sec;
6197         size_t old_sz, new_sz;
6198         const char *sec_name;
6199         int i, off_adj;
6200
6201         for_each_btf_ext_sec(ext_info, sec) {
6202                 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
6203                 if (!sec_name)
6204                         return -EINVAL;
6205                 if (strcmp(sec_name, prog->sec_name) != 0)
6206                         continue;
6207
6208                 for_each_btf_ext_rec(ext_info, sec, i, rec) {
6209                         __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ;
6210
6211                         if (insn_off < prog->sec_insn_off)
6212                                 continue;
6213                         if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt)
6214                                 break;
6215
6216                         if (!copy_start)
6217                                 copy_start = rec;
6218                         copy_end = rec + ext_info->rec_size;
6219                 }
6220
6221                 if (!copy_start)
6222                         return -ENOENT;
6223
6224                 /* append func/line info of a given (sub-)program to the main
6225                  * program func/line info
6226                  */
6227                 old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size;
6228                 new_sz = old_sz + (copy_end - copy_start);
6229                 new_prog_info = realloc(*prog_info, new_sz);
6230                 if (!new_prog_info)
6231                         return -ENOMEM;
6232                 *prog_info = new_prog_info;
6233                 *prog_rec_cnt = new_sz / ext_info->rec_size;
6234                 memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start);
6235
6236                 /* Kernel instruction offsets are in units of 8-byte
6237                  * instructions, while .BTF.ext instruction offsets generated
6238                  * by Clang are in units of bytes. So convert Clang offsets
6239                  * into kernel offsets and adjust offset according to program
6240                  * relocated position.
6241                  */
6242                 off_adj = prog->sub_insn_off - prog->sec_insn_off;
6243                 rec = new_prog_info + old_sz;
6244                 rec_end = new_prog_info + new_sz;
6245                 for (; rec < rec_end; rec += ext_info->rec_size) {
6246                         __u32 *insn_off = rec;
6247
6248                         *insn_off = *insn_off / BPF_INSN_SZ + off_adj;
6249                 }
6250                 *prog_rec_sz = ext_info->rec_size;
6251                 return 0;
6252         }
6253
6254         return -ENOENT;
6255 }
6256
6257 static int
6258 reloc_prog_func_and_line_info(const struct bpf_object *obj,
6259                               struct bpf_program *main_prog,
6260                               const struct bpf_program *prog)
6261 {
6262         int err;
6263
6264         /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't
6265          * supprot func/line info
6266          */
6267         if (!obj->btf_ext || !kernel_supports(FEAT_BTF_FUNC))
6268                 return 0;
6269
6270         /* only attempt func info relocation if main program's func_info
6271          * relocation was successful
6272          */
6273         if (main_prog != prog && !main_prog->func_info)
6274                 goto line_info;
6275
6276         err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info,
6277                                        &main_prog->func_info,
6278                                        &main_prog->func_info_cnt,
6279                                        &main_prog->func_info_rec_size);
6280         if (err) {
6281                 if (err != -ENOENT) {
6282                         pr_warn("prog '%s': error relocating .BTF.ext function info: %d\n",
6283                                 prog->name, err);
6284                         return err;
6285                 }
6286                 if (main_prog->func_info) {
6287                         /*
6288                          * Some info has already been found but has problem
6289                          * in the last btf_ext reloc. Must have to error out.
6290                          */
6291                         pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name);
6292                         return err;
6293                 }
6294                 /* Have problem loading the very first info. Ignore the rest. */
6295                 pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n",
6296                         prog->name);
6297         }
6298
6299 line_info:
6300         /* don't relocate line info if main program's relocation failed */
6301         if (main_prog != prog && !main_prog->line_info)
6302                 return 0;
6303
6304         err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info,
6305                                        &main_prog->line_info,
6306                                        &main_prog->line_info_cnt,
6307                                        &main_prog->line_info_rec_size);
6308         if (err) {
6309                 if (err != -ENOENT) {
6310                         pr_warn("prog '%s': error relocating .BTF.ext line info: %d\n",
6311                                 prog->name, err);
6312                         return err;
6313                 }
6314                 if (main_prog->line_info) {
6315                         /*
6316                          * Some info has already been found but has problem
6317                          * in the last btf_ext reloc. Must have to error out.
6318                          */
6319                         pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name);
6320                         return err;
6321                 }
6322                 /* Have problem loading the very first info. Ignore the rest. */
6323                 pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n",
6324                         prog->name);
6325         }
6326         return 0;
6327 }
6328
6329 static int cmp_relo_by_insn_idx(const void *key, const void *elem)
6330 {
6331         size_t insn_idx = *(const size_t *)key;
6332         const struct reloc_desc *relo = elem;
6333
6334         if (insn_idx == relo->insn_idx)
6335                 return 0;
6336         return insn_idx < relo->insn_idx ? -1 : 1;
6337 }
6338
6339 static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx)
6340 {
6341         return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc,
6342                        sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx);
6343 }
6344
6345 static int
6346 bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog,
6347                        struct bpf_program *prog)
6348 {
6349         size_t sub_insn_idx, insn_idx, new_cnt;
6350         struct bpf_program *subprog;
6351         struct bpf_insn *insns, *insn;
6352         struct reloc_desc *relo;
6353         int err;
6354
6355         err = reloc_prog_func_and_line_info(obj, main_prog, prog);
6356         if (err)
6357                 return err;
6358
6359         for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) {
6360                 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6361                 if (!insn_is_subprog_call(insn))
6362                         continue;
6363
6364                 relo = find_prog_insn_relo(prog, insn_idx);
6365                 if (relo && relo->type != RELO_CALL) {
6366                         pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n",
6367                                 prog->name, insn_idx, relo->type);
6368                         return -LIBBPF_ERRNO__RELOC;
6369                 }
6370                 if (relo) {
6371                         /* sub-program instruction index is a combination of
6372                          * an offset of a symbol pointed to by relocation and
6373                          * call instruction's imm field; for global functions,
6374                          * call always has imm = -1, but for static functions
6375                          * relocation is against STT_SECTION and insn->imm
6376                          * points to a start of a static function
6377                          */
6378                         sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1;
6379                 } else {
6380                         /* if subprogram call is to a static function within
6381                          * the same ELF section, there won't be any relocation
6382                          * emitted, but it also means there is no additional
6383                          * offset necessary, insns->imm is relative to
6384                          * instruction's original position within the section
6385                          */
6386                         sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1;
6387                 }
6388
6389                 /* we enforce that sub-programs should be in .text section */
6390                 subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx);
6391                 if (!subprog) {
6392                         pr_warn("prog '%s': no .text section found yet sub-program call exists\n",
6393                                 prog->name);
6394                         return -LIBBPF_ERRNO__RELOC;
6395                 }
6396
6397                 /* if it's the first call instruction calling into this
6398                  * subprogram (meaning this subprog hasn't been processed
6399                  * yet) within the context of current main program:
6400                  *   - append it at the end of main program's instructions blog;
6401                  *   - process is recursively, while current program is put on hold;
6402                  *   - if that subprogram calls some other not yet processes
6403                  *   subprogram, same thing will happen recursively until
6404                  *   there are no more unprocesses subprograms left to append
6405                  *   and relocate.
6406                  */
6407                 if (subprog->sub_insn_off == 0) {
6408                         subprog->sub_insn_off = main_prog->insns_cnt;
6409
6410                         new_cnt = main_prog->insns_cnt + subprog->insns_cnt;
6411                         insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns));
6412                         if (!insns) {
6413                                 pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name);
6414                                 return -ENOMEM;
6415                         }
6416                         main_prog->insns = insns;
6417                         main_prog->insns_cnt = new_cnt;
6418
6419                         memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns,
6420                                subprog->insns_cnt * sizeof(*insns));
6421
6422                         pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n",
6423                                  main_prog->name, subprog->insns_cnt, subprog->name);
6424
6425                         err = bpf_object__reloc_code(obj, main_prog, subprog);
6426                         if (err)
6427                                 return err;
6428                 }
6429
6430                 /* main_prog->insns memory could have been re-allocated, so
6431                  * calculate pointer again
6432                  */
6433                 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6434                 /* calculate correct instruction position within current main
6435                  * prog; each main prog can have a different set of
6436                  * subprograms appended (potentially in different order as
6437                  * well), so position of any subprog can be different for
6438                  * different main programs */
6439                 insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1;
6440
6441                 if (relo)
6442                         relo->processed = true;
6443
6444                 pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n",
6445                          prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off);
6446         }
6447
6448         return 0;
6449 }
6450
6451 /*
6452  * Relocate sub-program calls.
6453  *
6454  * Algorithm operates as follows. Each entry-point BPF program (referred to as
6455  * main prog) is processed separately. For each subprog (non-entry functions,
6456  * that can be called from either entry progs or other subprogs) gets their
6457  * sub_insn_off reset to zero. This serves as indicator that this subprogram
6458  * hasn't been yet appended and relocated within current main prog. Once its
6459  * relocated, sub_insn_off will point at the position within current main prog
6460  * where given subprog was appended. This will further be used to relocate all
6461  * the call instructions jumping into this subprog.
6462  *
6463  * We start with main program and process all call instructions. If the call
6464  * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off
6465  * is zero), subprog instructions are appended at the end of main program's
6466  * instruction array. Then main program is "put on hold" while we recursively
6467  * process newly appended subprogram. If that subprogram calls into another
6468  * subprogram that hasn't been appended, new subprogram is appended again to
6469  * the *main* prog's instructions (subprog's instructions are always left
6470  * untouched, as they need to be in unmodified state for subsequent main progs
6471  * and subprog instructions are always sent only as part of a main prog) and
6472  * the process continues recursively. Once all the subprogs called from a main
6473  * prog or any of its subprogs are appended (and relocated), all their
6474  * positions within finalized instructions array are known, so it's easy to
6475  * rewrite call instructions with correct relative offsets, corresponding to
6476  * desired target subprog.
6477  *
6478  * Its important to realize that some subprogs might not be called from some
6479  * main prog and any of its called/used subprogs. Those will keep their
6480  * subprog->sub_insn_off as zero at all times and won't be appended to current
6481  * main prog and won't be relocated within the context of current main prog.
6482  * They might still be used from other main progs later.
6483  *
6484  * Visually this process can be shown as below. Suppose we have two main
6485  * programs mainA and mainB and BPF object contains three subprogs: subA,
6486  * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and
6487  * subC both call subB:
6488  *
6489  *        +--------+ +-------+
6490  *        |        v v       |
6491  *     +--+---+ +--+-+-+ +---+--+
6492  *     | subA | | subB | | subC |
6493  *     +--+---+ +------+ +---+--+
6494  *        ^                  ^
6495  *        |                  |
6496  *    +---+-------+   +------+----+
6497  *    |   mainA   |   |   mainB   |
6498  *    +-----------+   +-----------+
6499  *
6500  * We'll start relocating mainA, will find subA, append it and start
6501  * processing sub A recursively:
6502  *
6503  *    +-----------+------+
6504  *    |   mainA   | subA |
6505  *    +-----------+------+
6506  *
6507  * At this point we notice that subB is used from subA, so we append it and
6508  * relocate (there are no further subcalls from subB):
6509  *
6510  *    +-----------+------+------+
6511  *    |   mainA   | subA | subB |
6512  *    +-----------+------+------+
6513  *
6514  * At this point, we relocate subA calls, then go one level up and finish with
6515  * relocatin mainA calls. mainA is done.
6516  *
6517  * For mainB process is similar but results in different order. We start with
6518  * mainB and skip subA and subB, as mainB never calls them (at least
6519  * directly), but we see subC is needed, so we append and start processing it:
6520  *
6521  *    +-----------+------+
6522  *    |   mainB   | subC |
6523  *    +-----------+------+
6524  * Now we see subC needs subB, so we go back to it, append and relocate it:
6525  *
6526  *    +-----------+------+------+
6527  *    |   mainB   | subC | subB |
6528  *    +-----------+------+------+
6529  *
6530  * At this point we unwind recursion, relocate calls in subC, then in mainB.
6531  */
6532 static int
6533 bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog)
6534 {
6535         struct bpf_program *subprog;
6536         int i, j, err;
6537
6538         /* mark all subprogs as not relocated (yet) within the context of
6539          * current main program
6540          */
6541         for (i = 0; i < obj->nr_programs; i++) {
6542                 subprog = &obj->programs[i];
6543                 if (!prog_is_subprog(obj, subprog))
6544                         continue;
6545
6546                 subprog->sub_insn_off = 0;
6547                 for (j = 0; j < subprog->nr_reloc; j++)
6548                         if (subprog->reloc_desc[j].type == RELO_CALL)
6549                                 subprog->reloc_desc[j].processed = false;
6550         }
6551
6552         err = bpf_object__reloc_code(obj, prog, prog);
6553         if (err)
6554                 return err;
6555
6556
6557         return 0;
6558 }
6559
6560 static int
6561 bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
6562 {
6563         struct bpf_program *prog;
6564         size_t i;
6565         int err;
6566
6567         if (obj->btf_ext) {
6568                 err = bpf_object__relocate_core(obj, targ_btf_path);
6569                 if (err) {
6570                         pr_warn("failed to perform CO-RE relocations: %d\n",
6571                                 err);
6572                         return err;
6573                 }
6574         }
6575         /* relocate data references first for all programs and sub-programs,
6576          * as they don't change relative to code locations, so subsequent
6577          * subprogram processing won't need to re-calculate any of them
6578          */
6579         for (i = 0; i < obj->nr_programs; i++) {
6580                 prog = &obj->programs[i];
6581                 err = bpf_object__relocate_data(obj, prog);
6582                 if (err) {
6583                         pr_warn("prog '%s': failed to relocate data references: %d\n",
6584                                 prog->name, err);
6585                         return err;
6586                 }
6587         }
6588         /* now relocate subprogram calls and append used subprograms to main
6589          * programs; each copy of subprogram code needs to be relocated
6590          * differently for each main program, because its code location might
6591          * have changed
6592          */
6593         for (i = 0; i < obj->nr_programs; i++) {
6594                 prog = &obj->programs[i];
6595                 /* sub-program's sub-calls are relocated within the context of
6596                  * its main program only
6597                  */
6598                 if (prog_is_subprog(obj, prog))
6599                         continue;
6600
6601                 err = bpf_object__relocate_calls(obj, prog);
6602                 if (err) {
6603                         pr_warn("prog '%s': failed to relocate calls: %d\n",
6604                                 prog->name, err);
6605                         return err;
6606                 }
6607         }
6608         /* free up relocation descriptors */
6609         for (i = 0; i < obj->nr_programs; i++) {
6610                 prog = &obj->programs[i];
6611                 zfree(&prog->reloc_desc);
6612                 prog->nr_reloc = 0;
6613         }
6614         return 0;
6615 }
6616
6617 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
6618                                             GElf_Shdr *shdr, Elf_Data *data);
6619
6620 static int bpf_object__collect_map_relos(struct bpf_object *obj,
6621                                          GElf_Shdr *shdr, Elf_Data *data)
6622 {
6623         const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *);
6624         int i, j, nrels, new_sz;
6625         const struct btf_var_secinfo *vi = NULL;
6626         const struct btf_type *sec, *var, *def;
6627         struct bpf_map *map = NULL, *targ_map;
6628         const struct btf_member *member;
6629         const char *name, *mname;
6630         Elf_Data *symbols;
6631         unsigned int moff;
6632         GElf_Sym sym;
6633         GElf_Rel rel;
6634         void *tmp;
6635
6636         if (!obj->efile.btf_maps_sec_btf_id || !obj->btf)
6637                 return -EINVAL;
6638         sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id);
6639         if (!sec)
6640                 return -EINVAL;
6641
6642         symbols = obj->efile.symbols;
6643         nrels = shdr->sh_size / shdr->sh_entsize;
6644         for (i = 0; i < nrels; i++) {
6645                 if (!gelf_getrel(data, i, &rel)) {
6646                         pr_warn(".maps relo #%d: failed to get ELF relo\n", i);
6647                         return -LIBBPF_ERRNO__FORMAT;
6648                 }
6649                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
6650                         pr_warn(".maps relo #%d: symbol %zx not found\n",
6651                                 i, (size_t)GELF_R_SYM(rel.r_info));
6652                         return -LIBBPF_ERRNO__FORMAT;
6653                 }
6654                 name = elf_sym_str(obj, sym.st_name) ?: "<?>";
6655                 if (sym.st_shndx != obj->efile.btf_maps_shndx) {
6656                         pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n",
6657                                 i, name);
6658                         return -LIBBPF_ERRNO__RELOC;
6659                 }
6660
6661                 pr_debug(".maps relo #%d: for %zd value %zd rel.r_offset %zu name %d ('%s')\n",
6662                          i, (ssize_t)(rel.r_info >> 32), (size_t)sym.st_value,
6663                          (size_t)rel.r_offset, sym.st_name, name);
6664
6665                 for (j = 0; j < obj->nr_maps; j++) {
6666                         map = &obj->maps[j];
6667                         if (map->sec_idx != obj->efile.btf_maps_shndx)
6668                                 continue;
6669
6670                         vi = btf_var_secinfos(sec) + map->btf_var_idx;
6671                         if (vi->offset <= rel.r_offset &&
6672                             rel.r_offset + bpf_ptr_sz <= vi->offset + vi->size)
6673                                 break;
6674                 }
6675                 if (j == obj->nr_maps) {
6676                         pr_warn(".maps relo #%d: cannot find map '%s' at rel.r_offset %zu\n",
6677                                 i, name, (size_t)rel.r_offset);
6678                         return -EINVAL;
6679                 }
6680
6681                 if (!bpf_map_type__is_map_in_map(map->def.type))
6682                         return -EINVAL;
6683                 if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS &&
6684                     map->def.key_size != sizeof(int)) {
6685                         pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n",
6686                                 i, map->name, sizeof(int));
6687                         return -EINVAL;
6688                 }
6689
6690                 targ_map = bpf_object__find_map_by_name(obj, name);
6691                 if (!targ_map)
6692                         return -ESRCH;
6693
6694                 var = btf__type_by_id(obj->btf, vi->type);
6695                 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
6696                 if (btf_vlen(def) == 0)
6697                         return -EINVAL;
6698                 member = btf_members(def) + btf_vlen(def) - 1;
6699                 mname = btf__name_by_offset(obj->btf, member->name_off);
6700                 if (strcmp(mname, "values"))
6701                         return -EINVAL;
6702
6703                 moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8;
6704                 if (rel.r_offset - vi->offset < moff)
6705                         return -EINVAL;
6706
6707                 moff = rel.r_offset - vi->offset - moff;
6708                 /* here we use BPF pointer size, which is always 64 bit, as we
6709                  * are parsing ELF that was built for BPF target
6710                  */
6711                 if (moff % bpf_ptr_sz)
6712                         return -EINVAL;
6713                 moff /= bpf_ptr_sz;
6714                 if (moff >= map->init_slots_sz) {
6715                         new_sz = moff + 1;
6716                         tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz);
6717                         if (!tmp)
6718                                 return -ENOMEM;
6719                         map->init_slots = tmp;
6720                         memset(map->init_slots + map->init_slots_sz, 0,
6721                                (new_sz - map->init_slots_sz) * host_ptr_sz);
6722                         map->init_slots_sz = new_sz;
6723                 }
6724                 map->init_slots[moff] = targ_map;
6725
6726                 pr_debug(".maps relo #%d: map '%s' slot [%d] points to map '%s'\n",
6727                          i, map->name, moff, name);
6728         }
6729
6730         return 0;
6731 }
6732
6733 static int cmp_relocs(const void *_a, const void *_b)
6734 {
6735         const struct reloc_desc *a = _a;
6736         const struct reloc_desc *b = _b;
6737
6738         if (a->insn_idx != b->insn_idx)
6739                 return a->insn_idx < b->insn_idx ? -1 : 1;
6740
6741         /* no two relocations should have the same insn_idx, but ... */
6742         if (a->type != b->type)
6743                 return a->type < b->type ? -1 : 1;
6744
6745         return 0;
6746 }
6747
6748 static int bpf_object__collect_relos(struct bpf_object *obj)
6749 {
6750         int i, err;
6751
6752         for (i = 0; i < obj->efile.nr_reloc_sects; i++) {
6753                 GElf_Shdr *shdr = &obj->efile.reloc_sects[i].shdr;
6754                 Elf_Data *data = obj->efile.reloc_sects[i].data;
6755                 int idx = shdr->sh_info;
6756
6757                 if (shdr->sh_type != SHT_REL) {
6758                         pr_warn("internal error at %d\n", __LINE__);
6759                         return -LIBBPF_ERRNO__INTERNAL;
6760                 }
6761
6762                 if (idx == obj->efile.st_ops_shndx)
6763                         err = bpf_object__collect_st_ops_relos(obj, shdr, data);
6764                 else if (idx == obj->efile.btf_maps_shndx)
6765                         err = bpf_object__collect_map_relos(obj, shdr, data);
6766                 else
6767                         err = bpf_object__collect_prog_relos(obj, shdr, data);
6768                 if (err)
6769                         return err;
6770         }
6771
6772         for (i = 0; i < obj->nr_programs; i++) {
6773                 struct bpf_program *p = &obj->programs[i];
6774
6775                 if (!p->nr_reloc)
6776                         continue;
6777
6778                 qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs);
6779         }
6780         return 0;
6781 }
6782
6783 static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id)
6784 {
6785         if (BPF_CLASS(insn->code) == BPF_JMP &&
6786             BPF_OP(insn->code) == BPF_CALL &&
6787             BPF_SRC(insn->code) == BPF_K &&
6788             insn->src_reg == 0 &&
6789             insn->dst_reg == 0) {
6790                     *func_id = insn->imm;
6791                     return true;
6792         }
6793         return false;
6794 }
6795
6796 static int bpf_object__sanitize_prog(struct bpf_object* obj, struct bpf_program *prog)
6797 {
6798         struct bpf_insn *insn = prog->insns;
6799         enum bpf_func_id func_id;
6800         int i;
6801
6802         for (i = 0; i < prog->insns_cnt; i++, insn++) {
6803                 if (!insn_is_helper_call(insn, &func_id))
6804                         continue;
6805
6806                 /* on kernels that don't yet support
6807                  * bpf_probe_read_{kernel,user}[_str] helpers, fall back
6808                  * to bpf_probe_read() which works well for old kernels
6809                  */
6810                 switch (func_id) {
6811                 case BPF_FUNC_probe_read_kernel:
6812                 case BPF_FUNC_probe_read_user:
6813                         if (!kernel_supports(FEAT_PROBE_READ_KERN))
6814                                 insn->imm = BPF_FUNC_probe_read;
6815                         break;
6816                 case BPF_FUNC_probe_read_kernel_str:
6817                 case BPF_FUNC_probe_read_user_str:
6818                         if (!kernel_supports(FEAT_PROBE_READ_KERN))
6819                                 insn->imm = BPF_FUNC_probe_read_str;
6820                         break;
6821                 default:
6822                         break;
6823                 }
6824         }
6825         return 0;
6826 }
6827
6828 static int
6829 load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
6830              char *license, __u32 kern_version, int *pfd)
6831 {
6832         struct bpf_prog_load_params load_attr = {};
6833         char *cp, errmsg[STRERR_BUFSIZE];
6834         size_t log_buf_size = 0;
6835         char *log_buf = NULL;
6836         int btf_fd, ret;
6837
6838         if (prog->type == BPF_PROG_TYPE_UNSPEC) {
6839                 /*
6840                  * The program type must be set.  Most likely we couldn't find a proper
6841                  * section definition at load time, and thus we didn't infer the type.
6842                  */
6843                 pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n",
6844                         prog->name, prog->sec_name);
6845                 return -EINVAL;
6846         }
6847
6848         if (!insns || !insns_cnt)
6849                 return -EINVAL;
6850
6851         load_attr.prog_type = prog->type;
6852         /* old kernels might not support specifying expected_attach_type */
6853         if (!kernel_supports(FEAT_EXP_ATTACH_TYPE) && prog->sec_def &&
6854             prog->sec_def->is_exp_attach_type_optional)
6855                 load_attr.expected_attach_type = 0;
6856         else
6857                 load_attr.expected_attach_type = prog->expected_attach_type;
6858         if (kernel_supports(FEAT_PROG_NAME))
6859                 load_attr.name = prog->name;
6860         load_attr.insns = insns;
6861         load_attr.insn_cnt = insns_cnt;
6862         load_attr.license = license;
6863         load_attr.attach_btf_id = prog->attach_btf_id;
6864         if (prog->attach_prog_fd)
6865                 load_attr.attach_prog_fd = prog->attach_prog_fd;
6866         else
6867                 load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd;
6868         load_attr.attach_btf_id = prog->attach_btf_id;
6869         load_attr.kern_version = kern_version;
6870         load_attr.prog_ifindex = prog->prog_ifindex;
6871
6872         /* specify func_info/line_info only if kernel supports them */
6873         btf_fd = bpf_object__btf_fd(prog->obj);
6874         if (btf_fd >= 0 && kernel_supports(FEAT_BTF_FUNC)) {
6875                 load_attr.prog_btf_fd = btf_fd;
6876                 load_attr.func_info = prog->func_info;
6877                 load_attr.func_info_rec_size = prog->func_info_rec_size;
6878                 load_attr.func_info_cnt = prog->func_info_cnt;
6879                 load_attr.line_info = prog->line_info;
6880                 load_attr.line_info_rec_size = prog->line_info_rec_size;
6881                 load_attr.line_info_cnt = prog->line_info_cnt;
6882         }
6883         load_attr.log_level = prog->log_level;
6884         load_attr.prog_flags = prog->prog_flags;
6885
6886 retry_load:
6887         if (log_buf_size) {
6888                 log_buf = malloc(log_buf_size);
6889                 if (!log_buf)
6890                         return -ENOMEM;
6891
6892                 *log_buf = 0;
6893         }
6894
6895         load_attr.log_buf = log_buf;
6896         load_attr.log_buf_sz = log_buf_size;
6897         ret = libbpf__bpf_prog_load(&load_attr);
6898
6899         if (ret >= 0) {
6900                 if (log_buf && load_attr.log_level)
6901                         pr_debug("verifier log:\n%s", log_buf);
6902
6903                 if (prog->obj->rodata_map_idx >= 0 &&
6904                     kernel_supports(FEAT_PROG_BIND_MAP)) {
6905                         struct bpf_map *rodata_map =
6906                                 &prog->obj->maps[prog->obj->rodata_map_idx];
6907
6908                         if (bpf_prog_bind_map(ret, bpf_map__fd(rodata_map), NULL)) {
6909                                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
6910                                 pr_warn("prog '%s': failed to bind .rodata map: %s\n",
6911                                         prog->name, cp);
6912                                 /* Don't fail hard if can't bind rodata. */
6913                         }
6914                 }
6915
6916                 *pfd = ret;
6917                 ret = 0;
6918                 goto out;
6919         }
6920
6921         if (!log_buf || errno == ENOSPC) {
6922                 log_buf_size = max((size_t)BPF_LOG_BUF_SIZE,
6923                                    log_buf_size << 1);
6924
6925                 free(log_buf);
6926                 goto retry_load;
6927         }
6928         ret = errno ? -errno : -LIBBPF_ERRNO__LOAD;
6929         cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
6930         pr_warn("load bpf program failed: %s\n", cp);
6931         pr_perm_msg(ret);
6932
6933         if (log_buf && log_buf[0] != '\0') {
6934                 ret = -LIBBPF_ERRNO__VERIFY;
6935                 pr_warn("-- BEGIN DUMP LOG ---\n");
6936                 pr_warn("\n%s\n", log_buf);
6937                 pr_warn("-- END LOG --\n");
6938         } else if (load_attr.insn_cnt >= BPF_MAXINSNS) {
6939                 pr_warn("Program too large (%zu insns), at most %d insns\n",
6940                         load_attr.insn_cnt, BPF_MAXINSNS);
6941                 ret = -LIBBPF_ERRNO__PROG2BIG;
6942         } else if (load_attr.prog_type != BPF_PROG_TYPE_KPROBE) {
6943                 /* Wrong program type? */
6944                 int fd;
6945
6946                 load_attr.prog_type = BPF_PROG_TYPE_KPROBE;
6947                 load_attr.expected_attach_type = 0;
6948                 load_attr.log_buf = NULL;
6949                 load_attr.log_buf_sz = 0;
6950                 fd = libbpf__bpf_prog_load(&load_attr);
6951                 if (fd >= 0) {
6952                         close(fd);
6953                         ret = -LIBBPF_ERRNO__PROGTYPE;
6954                         goto out;
6955                 }
6956         }
6957
6958 out:
6959         free(log_buf);
6960         return ret;
6961 }
6962
6963 static int libbpf_find_attach_btf_id(struct bpf_program *prog, int *btf_obj_fd, int *btf_type_id);
6964
6965 int bpf_program__load(struct bpf_program *prog, char *license, __u32 kern_ver)
6966 {
6967         int err = 0, fd, i;
6968
6969         if (prog->obj->loaded) {
6970                 pr_warn("prog '%s': can't load after object was loaded\n", prog->name);
6971                 return -EINVAL;
6972         }
6973
6974         if ((prog->type == BPF_PROG_TYPE_TRACING ||
6975              prog->type == BPF_PROG_TYPE_LSM ||
6976              prog->type == BPF_PROG_TYPE_EXT) && !prog->attach_btf_id) {
6977                 int btf_obj_fd = 0, btf_type_id = 0;
6978
6979                 err = libbpf_find_attach_btf_id(prog, &btf_obj_fd, &btf_type_id);
6980                 if (err)
6981                         return err;
6982
6983                 prog->attach_btf_obj_fd = btf_obj_fd;
6984                 prog->attach_btf_id = btf_type_id;
6985         }
6986
6987         if (prog->instances.nr < 0 || !prog->instances.fds) {
6988                 if (prog->preprocessor) {
6989                         pr_warn("Internal error: can't load program '%s'\n",
6990                                 prog->name);
6991                         return -LIBBPF_ERRNO__INTERNAL;
6992                 }
6993
6994                 prog->instances.fds = malloc(sizeof(int));
6995                 if (!prog->instances.fds) {
6996                         pr_warn("Not enough memory for BPF fds\n");
6997                         return -ENOMEM;
6998                 }
6999                 prog->instances.nr = 1;
7000                 prog->instances.fds[0] = -1;
7001         }
7002
7003         if (!prog->preprocessor) {
7004                 if (prog->instances.nr != 1) {
7005                         pr_warn("prog '%s': inconsistent nr(%d) != 1\n",
7006                                 prog->name, prog->instances.nr);
7007                 }
7008                 err = load_program(prog, prog->insns, prog->insns_cnt,
7009                                    license, kern_ver, &fd);
7010                 if (!err)
7011                         prog->instances.fds[0] = fd;
7012                 goto out;
7013         }
7014
7015         for (i = 0; i < prog->instances.nr; i++) {
7016                 struct bpf_prog_prep_result result;
7017                 bpf_program_prep_t preprocessor = prog->preprocessor;
7018
7019                 memset(&result, 0, sizeof(result));
7020                 err = preprocessor(prog, i, prog->insns,
7021                                    prog->insns_cnt, &result);
7022                 if (err) {
7023                         pr_warn("Preprocessing the %dth instance of program '%s' failed\n",
7024                                 i, prog->name);
7025                         goto out;
7026                 }
7027
7028                 if (!result.new_insn_ptr || !result.new_insn_cnt) {
7029                         pr_debug("Skip loading the %dth instance of program '%s'\n",
7030                                  i, prog->name);
7031                         prog->instances.fds[i] = -1;
7032                         if (result.pfd)
7033                                 *result.pfd = -1;
7034                         continue;
7035                 }
7036
7037                 err = load_program(prog, result.new_insn_ptr,
7038                                    result.new_insn_cnt, license, kern_ver, &fd);
7039                 if (err) {
7040                         pr_warn("Loading the %dth instance of program '%s' failed\n",
7041                                 i, prog->name);
7042                         goto out;
7043                 }
7044
7045                 if (result.pfd)
7046                         *result.pfd = fd;
7047                 prog->instances.fds[i] = fd;
7048         }
7049 out:
7050         if (err)
7051                 pr_warn("failed to load program '%s'\n", prog->name);
7052         zfree(&prog->insns);
7053         prog->insns_cnt = 0;
7054         return err;
7055 }
7056
7057 static int
7058 bpf_object__load_progs(struct bpf_object *obj, int log_level)
7059 {
7060         struct bpf_program *prog;
7061         size_t i;
7062         int err;
7063
7064         for (i = 0; i < obj->nr_programs; i++) {
7065                 prog = &obj->programs[i];
7066                 err = bpf_object__sanitize_prog(obj, prog);
7067                 if (err)
7068                         return err;
7069         }
7070
7071         for (i = 0; i < obj->nr_programs; i++) {
7072                 prog = &obj->programs[i];
7073                 if (prog_is_subprog(obj, prog))
7074                         continue;
7075                 if (!prog->load) {
7076                         pr_debug("prog '%s': skipped loading\n", prog->name);
7077                         continue;
7078                 }
7079                 prog->log_level |= log_level;
7080                 err = bpf_program__load(prog, obj->license, obj->kern_version);
7081                 if (err)
7082                         return err;
7083         }
7084         return 0;
7085 }
7086
7087 static const struct bpf_sec_def *find_sec_def(const char *sec_name);
7088
7089 static struct bpf_object *
7090 __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
7091                    const struct bpf_object_open_opts *opts)
7092 {
7093         const char *obj_name, *kconfig;
7094         struct bpf_program *prog;
7095         struct bpf_object *obj;
7096         char tmp_name[64];
7097         int err;
7098
7099         if (elf_version(EV_CURRENT) == EV_NONE) {
7100                 pr_warn("failed to init libelf for %s\n",
7101                         path ? : "(mem buf)");
7102                 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
7103         }
7104
7105         if (!OPTS_VALID(opts, bpf_object_open_opts))
7106                 return ERR_PTR(-EINVAL);
7107
7108         obj_name = OPTS_GET(opts, object_name, NULL);
7109         if (obj_buf) {
7110                 if (!obj_name) {
7111                         snprintf(tmp_name, sizeof(tmp_name), "%lx-%lx",
7112                                  (unsigned long)obj_buf,
7113                                  (unsigned long)obj_buf_sz);
7114                         obj_name = tmp_name;
7115                 }
7116                 path = obj_name;
7117                 pr_debug("loading object '%s' from buffer\n", obj_name);
7118         }
7119
7120         obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
7121         if (IS_ERR(obj))
7122                 return obj;
7123
7124         kconfig = OPTS_GET(opts, kconfig, NULL);
7125         if (kconfig) {
7126                 obj->kconfig = strdup(kconfig);
7127                 if (!obj->kconfig)
7128                         return ERR_PTR(-ENOMEM);
7129         }
7130
7131         err = bpf_object__elf_init(obj);
7132         err = err ? : bpf_object__check_endianness(obj);
7133         err = err ? : bpf_object__elf_collect(obj);
7134         err = err ? : bpf_object__collect_externs(obj);
7135         err = err ? : bpf_object__finalize_btf(obj);
7136         err = err ? : bpf_object__init_maps(obj, opts);
7137         err = err ? : bpf_object__collect_relos(obj);
7138         if (err)
7139                 goto out;
7140         bpf_object__elf_finish(obj);
7141
7142         bpf_object__for_each_program(prog, obj) {
7143                 prog->sec_def = find_sec_def(prog->sec_name);
7144                 if (!prog->sec_def) {
7145                         /* couldn't guess, but user might manually specify */
7146                         pr_debug("prog '%s': unrecognized ELF section name '%s'\n",
7147                                 prog->name, prog->sec_name);
7148                         continue;
7149                 }
7150
7151                 if (prog->sec_def->is_sleepable)
7152                         prog->prog_flags |= BPF_F_SLEEPABLE;
7153                 bpf_program__set_type(prog, prog->sec_def->prog_type);
7154                 bpf_program__set_expected_attach_type(prog,
7155                                 prog->sec_def->expected_attach_type);
7156
7157                 if (prog->sec_def->prog_type == BPF_PROG_TYPE_TRACING ||
7158                     prog->sec_def->prog_type == BPF_PROG_TYPE_EXT)
7159                         prog->attach_prog_fd = OPTS_GET(opts, attach_prog_fd, 0);
7160         }
7161
7162         return obj;
7163 out:
7164         bpf_object__close(obj);
7165         return ERR_PTR(err);
7166 }
7167
7168 static struct bpf_object *
7169 __bpf_object__open_xattr(struct bpf_object_open_attr *attr, int flags)
7170 {
7171         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
7172                 .relaxed_maps = flags & MAPS_RELAX_COMPAT,
7173         );
7174
7175         /* param validation */
7176         if (!attr->file)
7177                 return NULL;
7178
7179         pr_debug("loading %s\n", attr->file);
7180         return __bpf_object__open(attr->file, NULL, 0, &opts);
7181 }
7182
7183 struct bpf_object *bpf_object__open_xattr(struct bpf_object_open_attr *attr)
7184 {
7185         return __bpf_object__open_xattr(attr, 0);
7186 }
7187
7188 struct bpf_object *bpf_object__open(const char *path)
7189 {
7190         struct bpf_object_open_attr attr = {
7191                 .file           = path,
7192                 .prog_type      = BPF_PROG_TYPE_UNSPEC,
7193         };
7194
7195         return bpf_object__open_xattr(&attr);
7196 }
7197
7198 struct bpf_object *
7199 bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
7200 {
7201         if (!path)
7202                 return ERR_PTR(-EINVAL);
7203
7204         pr_debug("loading %s\n", path);
7205
7206         return __bpf_object__open(path, NULL, 0, opts);
7207 }
7208
7209 struct bpf_object *
7210 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
7211                      const struct bpf_object_open_opts *opts)
7212 {
7213         if (!obj_buf || obj_buf_sz == 0)
7214                 return ERR_PTR(-EINVAL);
7215
7216         return __bpf_object__open(NULL, obj_buf, obj_buf_sz, opts);
7217 }
7218
7219 struct bpf_object *
7220 bpf_object__open_buffer(const void *obj_buf, size_t obj_buf_sz,
7221                         const char *name)
7222 {
7223         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, opts,
7224                 .object_name = name,
7225                 /* wrong default, but backwards-compatible */
7226                 .relaxed_maps = true,
7227         );
7228
7229         /* returning NULL is wrong, but backwards-compatible */
7230         if (!obj_buf || obj_buf_sz == 0)
7231                 return NULL;
7232
7233         return bpf_object__open_mem(obj_buf, obj_buf_sz, &opts);
7234 }
7235
7236 int bpf_object__unload(struct bpf_object *obj)
7237 {
7238         size_t i;
7239
7240         if (!obj)
7241                 return -EINVAL;
7242
7243         for (i = 0; i < obj->nr_maps; i++) {
7244                 zclose(obj->maps[i].fd);
7245                 if (obj->maps[i].st_ops)
7246                         zfree(&obj->maps[i].st_ops->kern_vdata);
7247         }
7248
7249         for (i = 0; i < obj->nr_programs; i++)
7250                 bpf_program__unload(&obj->programs[i]);
7251
7252         return 0;
7253 }
7254
7255 static int bpf_object__sanitize_maps(struct bpf_object *obj)
7256 {
7257         struct bpf_map *m;
7258
7259         bpf_object__for_each_map(m, obj) {
7260                 if (!bpf_map__is_internal(m))
7261                         continue;
7262                 if (!kernel_supports(FEAT_GLOBAL_DATA)) {
7263                         pr_warn("kernel doesn't support global data\n");
7264                         return -ENOTSUP;
7265                 }
7266                 if (!kernel_supports(FEAT_ARRAY_MMAP))
7267                         m->def.map_flags ^= BPF_F_MMAPABLE;
7268         }
7269
7270         return 0;
7271 }
7272
7273 static int bpf_object__read_kallsyms_file(struct bpf_object *obj)
7274 {
7275         char sym_type, sym_name[500];
7276         unsigned long long sym_addr;
7277         struct extern_desc *ext;
7278         int ret, err = 0;
7279         FILE *f;
7280
7281         f = fopen("/proc/kallsyms", "r");
7282         if (!f) {
7283                 err = -errno;
7284                 pr_warn("failed to open /proc/kallsyms: %d\n", err);
7285                 return err;
7286         }
7287
7288         while (true) {
7289                 ret = fscanf(f, "%llx %c %499s%*[^\n]\n",
7290                              &sym_addr, &sym_type, sym_name);
7291                 if (ret == EOF && feof(f))
7292                         break;
7293                 if (ret != 3) {
7294                         pr_warn("failed to read kallsyms entry: %d\n", ret);
7295                         err = -EINVAL;
7296                         goto out;
7297                 }
7298
7299                 ext = find_extern_by_name(obj, sym_name);
7300                 if (!ext || ext->type != EXT_KSYM)
7301                         continue;
7302
7303                 if (ext->is_set && ext->ksym.addr != sym_addr) {
7304                         pr_warn("extern (ksym) '%s' resolution is ambiguous: 0x%llx or 0x%llx\n",
7305                                 sym_name, ext->ksym.addr, sym_addr);
7306                         err = -EINVAL;
7307                         goto out;
7308                 }
7309                 if (!ext->is_set) {
7310                         ext->is_set = true;
7311                         ext->ksym.addr = sym_addr;
7312                         pr_debug("extern (ksym) %s=0x%llx\n", sym_name, sym_addr);
7313                 }
7314         }
7315
7316 out:
7317         fclose(f);
7318         return err;
7319 }
7320
7321 static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj)
7322 {
7323         struct extern_desc *ext;
7324         struct btf *btf;
7325         int i, j, id, btf_fd, err;
7326
7327         for (i = 0; i < obj->nr_extern; i++) {
7328                 const struct btf_type *targ_var, *targ_type;
7329                 __u32 targ_type_id, local_type_id;
7330                 const char *targ_var_name;
7331                 int ret;
7332
7333                 ext = &obj->externs[i];
7334                 if (ext->type != EXT_KSYM || !ext->ksym.type_id)
7335                         continue;
7336
7337                 btf = obj->btf_vmlinux;
7338                 btf_fd = 0;
7339                 id = btf__find_by_name_kind(btf, ext->name, BTF_KIND_VAR);
7340                 if (id == -ENOENT) {
7341                         err = load_module_btfs(obj);
7342                         if (err)
7343                                 return err;
7344
7345                         for (j = 0; j < obj->btf_module_cnt; j++) {
7346                                 btf = obj->btf_modules[j].btf;
7347                                 /* we assume module BTF FD is always >0 */
7348                                 btf_fd = obj->btf_modules[j].fd;
7349                                 id = btf__find_by_name_kind(btf, ext->name, BTF_KIND_VAR);
7350                                 if (id != -ENOENT)
7351                                         break;
7352                         }
7353                 }
7354                 if (id <= 0) {
7355                         pr_warn("extern (ksym) '%s': failed to find BTF ID in kernel BTF(s).\n",
7356                                 ext->name);
7357                         return -ESRCH;
7358                 }
7359
7360                 /* find local type_id */
7361                 local_type_id = ext->ksym.type_id;
7362
7363                 /* find target type_id */
7364                 targ_var = btf__type_by_id(btf, id);
7365                 targ_var_name = btf__name_by_offset(btf, targ_var->name_off);
7366                 targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id);
7367
7368                 ret = bpf_core_types_are_compat(obj->btf, local_type_id,
7369                                                 btf, targ_type_id);
7370                 if (ret <= 0) {
7371                         const struct btf_type *local_type;
7372                         const char *targ_name, *local_name;
7373
7374                         local_type = btf__type_by_id(obj->btf, local_type_id);
7375                         local_name = btf__name_by_offset(obj->btf, local_type->name_off);
7376                         targ_name = btf__name_by_offset(btf, targ_type->name_off);
7377
7378                         pr_warn("extern (ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n",
7379                                 ext->name, local_type_id,
7380                                 btf_kind_str(local_type), local_name, targ_type_id,
7381                                 btf_kind_str(targ_type), targ_name);
7382                         return -EINVAL;
7383                 }
7384
7385                 ext->is_set = true;
7386                 ext->ksym.kernel_btf_obj_fd = btf_fd;
7387                 ext->ksym.kernel_btf_id = id;
7388                 pr_debug("extern (ksym) '%s': resolved to [%d] %s %s\n",
7389                          ext->name, id, btf_kind_str(targ_var), targ_var_name);
7390         }
7391         return 0;
7392 }
7393
7394 static int bpf_object__resolve_externs(struct bpf_object *obj,
7395                                        const char *extra_kconfig)
7396 {
7397         bool need_config = false, need_kallsyms = false;
7398         bool need_vmlinux_btf = false;
7399         struct extern_desc *ext;
7400         void *kcfg_data = NULL;
7401         int err, i;
7402
7403         if (obj->nr_extern == 0)
7404                 return 0;
7405
7406         if (obj->kconfig_map_idx >= 0)
7407                 kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped;
7408
7409         for (i = 0; i < obj->nr_extern; i++) {
7410                 ext = &obj->externs[i];
7411
7412                 if (ext->type == EXT_KCFG &&
7413                     strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
7414                         void *ext_val = kcfg_data + ext->kcfg.data_off;
7415                         __u32 kver = get_kernel_version();
7416
7417                         if (!kver) {
7418                                 pr_warn("failed to get kernel version\n");
7419                                 return -EINVAL;
7420                         }
7421                         err = set_kcfg_value_num(ext, ext_val, kver);
7422                         if (err)
7423                                 return err;
7424                         pr_debug("extern (kcfg) %s=0x%x\n", ext->name, kver);
7425                 } else if (ext->type == EXT_KCFG &&
7426                            strncmp(ext->name, "CONFIG_", 7) == 0) {
7427                         need_config = true;
7428                 } else if (ext->type == EXT_KSYM) {
7429                         if (ext->ksym.type_id)
7430                                 need_vmlinux_btf = true;
7431                         else
7432                                 need_kallsyms = true;
7433                 } else {
7434                         pr_warn("unrecognized extern '%s'\n", ext->name);
7435                         return -EINVAL;
7436                 }
7437         }
7438         if (need_config && extra_kconfig) {
7439                 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data);
7440                 if (err)
7441                         return -EINVAL;
7442                 need_config = false;
7443                 for (i = 0; i < obj->nr_extern; i++) {
7444                         ext = &obj->externs[i];
7445                         if (ext->type == EXT_KCFG && !ext->is_set) {
7446                                 need_config = true;
7447                                 break;
7448                         }
7449                 }
7450         }
7451         if (need_config) {
7452                 err = bpf_object__read_kconfig_file(obj, kcfg_data);
7453                 if (err)
7454                         return -EINVAL;
7455         }
7456         if (need_kallsyms) {
7457                 err = bpf_object__read_kallsyms_file(obj);
7458                 if (err)
7459                         return -EINVAL;
7460         }
7461         if (need_vmlinux_btf) {
7462                 err = bpf_object__resolve_ksyms_btf_id(obj);
7463                 if (err)
7464                         return -EINVAL;
7465         }
7466         for (i = 0; i < obj->nr_extern; i++) {
7467                 ext = &obj->externs[i];
7468
7469                 if (!ext->is_set && !ext->is_weak) {
7470                         pr_warn("extern %s (strong) not resolved\n", ext->name);
7471                         return -ESRCH;
7472                 } else if (!ext->is_set) {
7473                         pr_debug("extern %s (weak) not resolved, defaulting to zero\n",
7474                                  ext->name);
7475                 }
7476         }
7477
7478         return 0;
7479 }
7480
7481 int bpf_object__load_xattr(struct bpf_object_load_attr *attr)
7482 {
7483         struct bpf_object *obj;
7484         int err, i;
7485
7486         if (!attr)
7487                 return -EINVAL;
7488         obj = attr->obj;
7489         if (!obj)
7490                 return -EINVAL;
7491
7492         if (obj->loaded) {
7493                 pr_warn("object '%s': load can't be attempted twice\n", obj->name);
7494                 return -EINVAL;
7495         }
7496
7497         err = bpf_object__probe_loading(obj);
7498         err = err ? : bpf_object__load_vmlinux_btf(obj, false);
7499         err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
7500         err = err ? : bpf_object__sanitize_and_load_btf(obj);
7501         err = err ? : bpf_object__sanitize_maps(obj);
7502         err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
7503         err = err ? : bpf_object__create_maps(obj);
7504         err = err ? : bpf_object__relocate(obj, attr->target_btf_path);
7505         err = err ? : bpf_object__load_progs(obj, attr->log_level);
7506
7507         /* clean up module BTFs */
7508         for (i = 0; i < obj->btf_module_cnt; i++) {
7509                 close(obj->btf_modules[i].fd);
7510                 btf__free(obj->btf_modules[i].btf);
7511                 free(obj->btf_modules[i].name);
7512         }
7513         free(obj->btf_modules);
7514
7515         /* clean up vmlinux BTF */
7516         btf__free(obj->btf_vmlinux);
7517         obj->btf_vmlinux = NULL;
7518
7519         obj->loaded = true; /* doesn't matter if successfully or not */
7520
7521         if (err)
7522                 goto out;
7523
7524         return 0;
7525 out:
7526         /* unpin any maps that were auto-pinned during load */
7527         for (i = 0; i < obj->nr_maps; i++)
7528                 if (obj->maps[i].pinned && !obj->maps[i].reused)
7529                         bpf_map__unpin(&obj->maps[i], NULL);
7530
7531         bpf_object__unload(obj);
7532         pr_warn("failed to load object '%s'\n", obj->path);
7533         return err;
7534 }
7535
7536 int bpf_object__load(struct bpf_object *obj)
7537 {
7538         struct bpf_object_load_attr attr = {
7539                 .obj = obj,
7540         };
7541
7542         return bpf_object__load_xattr(&attr);
7543 }
7544
7545 static int make_parent_dir(const char *path)
7546 {
7547         char *cp, errmsg[STRERR_BUFSIZE];
7548         char *dname, *dir;
7549         int err = 0;
7550
7551         dname = strdup(path);
7552         if (dname == NULL)
7553                 return -ENOMEM;
7554
7555         dir = dirname(dname);
7556         if (mkdir(dir, 0700) && errno != EEXIST)
7557                 err = -errno;
7558
7559         free(dname);
7560         if (err) {
7561                 cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
7562                 pr_warn("failed to mkdir %s: %s\n", path, cp);
7563         }
7564         return err;
7565 }
7566
7567 static int check_path(const char *path)
7568 {
7569         char *cp, errmsg[STRERR_BUFSIZE];
7570         struct statfs st_fs;
7571         char *dname, *dir;
7572         int err = 0;
7573
7574         if (path == NULL)
7575                 return -EINVAL;
7576
7577         dname = strdup(path);
7578         if (dname == NULL)
7579                 return -ENOMEM;
7580
7581         dir = dirname(dname);
7582         if (statfs(dir, &st_fs)) {
7583                 cp = libbpf_strerror_r(errno, errmsg, sizeof(errmsg));
7584                 pr_warn("failed to statfs %s: %s\n", dir, cp);
7585                 err = -errno;
7586         }
7587         free(dname);
7588
7589         if (!err && st_fs.f_type != BPF_FS_MAGIC) {
7590                 pr_warn("specified path %s is not on BPF FS\n", path);
7591                 err = -EINVAL;
7592         }
7593
7594         return err;
7595 }
7596
7597 int bpf_program__pin_instance(struct bpf_program *prog, const char *path,
7598                               int instance)
7599 {
7600         char *cp, errmsg[STRERR_BUFSIZE];
7601         int err;
7602
7603         err = make_parent_dir(path);
7604         if (err)
7605                 return err;
7606
7607         err = check_path(path);
7608         if (err)
7609                 return err;
7610
7611         if (prog == NULL) {
7612                 pr_warn("invalid program pointer\n");
7613                 return -EINVAL;
7614         }
7615
7616         if (instance < 0 || instance >= prog->instances.nr) {
7617                 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
7618                         instance, prog->name, prog->instances.nr);
7619                 return -EINVAL;
7620         }
7621
7622         if (bpf_obj_pin(prog->instances.fds[instance], path)) {
7623                 err = -errno;
7624                 cp = libbpf_strerror_r(err, errmsg, sizeof(errmsg));
7625                 pr_warn("failed to pin program: %s\n", cp);
7626                 return err;
7627         }
7628         pr_debug("pinned program '%s'\n", path);
7629
7630         return 0;
7631 }
7632
7633 int bpf_program__unpin_instance(struct bpf_program *prog, const char *path,
7634                                 int instance)
7635 {
7636         int err;
7637
7638         err = check_path(path);
7639         if (err)
7640                 return err;
7641
7642         if (prog == NULL) {
7643                 pr_warn("invalid program pointer\n");
7644                 return -EINVAL;
7645         }
7646
7647         if (instance < 0 || instance >= prog->instances.nr) {
7648                 pr_warn("invalid prog instance %d of prog %s (max %d)\n",
7649                         instance, prog->name, prog->instances.nr);
7650                 return -EINVAL;
7651         }
7652
7653         err = unlink(path);
7654         if (err != 0)
7655                 return -errno;
7656         pr_debug("unpinned program '%s'\n", path);
7657
7658         return 0;
7659 }
7660
7661 int bpf_program__pin(struct bpf_program *prog, const char *path)
7662 {
7663         int i, err;
7664
7665         err = make_parent_dir(path);
7666         if (err)
7667                 return err;
7668
7669         err = check_path(path);
7670         if (err)
7671                 return err;
7672
7673         if (prog == NULL) {
7674                 pr_warn("invalid program pointer\n");
7675                 return -EINVAL;
7676         }
7677
7678         if (prog->instances.nr <= 0) {
7679                 pr_warn("no instances of prog %s to pin\n", prog->name);
7680                 return -EINVAL;
7681         }
7682
7683         if (prog->instances.nr == 1) {
7684                 /* don't create subdirs when pinning single instance */
7685                 return bpf_program__pin_instance(prog, path, 0);
7686         }
7687
7688         for (i = 0; i < prog->instances.nr; i++) {
7689                 char buf[PATH_MAX];
7690                 int len;
7691
7692                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
7693                 if (len < 0) {
7694                         err = -EINVAL;
7695                         goto err_unpin;
7696                 } else if (len >= PATH_MAX) {
7697                         err = -ENAMETOOLONG;
7698                         goto err_unpin;
7699                 }
7700
7701                 err = bpf_program__pin_instance(prog, buf, i);
7702                 if (err)
7703                         goto err_unpin;
7704         }
7705
7706         return 0;
7707
7708 err_unpin:
7709         for (i = i - 1; i >= 0; i--) {
7710                 char buf[PATH_MAX];
7711                 int len;
7712
7713                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
7714                 if (len < 0)
7715                         continue;
7716                 else if (len >= PATH_MAX)
7717                         continue;
7718
7719                 bpf_program__unpin_instance(prog, buf, i);
7720         }
7721
7722         rmdir(path);
7723
7724         return err;
7725 }
7726
7727 int bpf_program__unpin(struct bpf_program *prog, const char *path)
7728 {
7729         int i, err;
7730
7731         err = check_path(path);
7732         if (err)
7733                 return err;
7734
7735         if (prog == NULL) {
7736                 pr_warn("invalid program pointer\n");
7737                 return -EINVAL;
7738         }
7739
7740         if (prog->instances.nr <= 0) {
7741                 pr_warn("no instances of prog %s to pin\n", prog->name);
7742                 return -EINVAL;
7743         }
7744
7745         if (prog->instances.nr == 1) {
7746                 /* don't create subdirs when pinning single instance */
7747                 return bpf_program__unpin_instance(prog, path, 0);
7748         }
7749
7750         for (i = 0; i < prog->instances.nr; i++) {
7751                 char buf[PATH_MAX];
7752                 int len;
7753
7754                 len = snprintf(buf, PATH_MAX, "%s/%d", path, i);
7755                 if (len < 0)
7756                         return -EINVAL;
7757                 else if (len >= PATH_MAX)
7758                         return -ENAMETOOLONG;
7759
7760                 err = bpf_program__unpin_instance(prog, buf, i);
7761                 if (err)
7762                         return err;
7763         }
7764
7765         err = rmdir(path);
7766         if (err)
7767                 return -errno;
7768
7769         return 0;
7770 }
7771
7772 int bpf_map__pin(struct bpf_map *map, const char *path)
7773 {
7774         char *cp, errmsg[STRERR_BUFSIZE];
7775         int err;
7776
7777         if (map == NULL) {
7778                 pr_warn("invalid map pointer\n");
7779                 return -EINVAL;
7780         }
7781
7782         if (map->pin_path) {
7783                 if (path && strcmp(path, map->pin_path)) {
7784                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
7785                                 bpf_map__name(map), map->pin_path, path);
7786                         return -EINVAL;
7787                 } else if (map->pinned) {
7788                         pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
7789                                  bpf_map__name(map), map->pin_path);
7790                         return 0;
7791                 }
7792         } else {
7793                 if (!path) {
7794                         pr_warn("missing a path to pin map '%s' at\n",
7795                                 bpf_map__name(map));
7796                         return -EINVAL;
7797                 } else if (map->pinned) {
7798                         pr_warn("map '%s' already pinned\n", bpf_map__name(map));
7799                         return -EEXIST;
7800                 }
7801
7802                 map->pin_path = strdup(path);
7803                 if (!map->pin_path) {
7804                         err = -errno;
7805                         goto out_err;
7806                 }
7807         }
7808
7809         err = make_parent_dir(map->pin_path);
7810         if (err)
7811                 return err;
7812
7813         err = check_path(map->pin_path);
7814         if (err)
7815                 return err;
7816
7817         if (bpf_obj_pin(map->fd, map->pin_path)) {
7818                 err = -errno;
7819                 goto out_err;
7820         }
7821
7822         map->pinned = true;
7823         pr_debug("pinned map '%s'\n", map->pin_path);
7824
7825         return 0;
7826
7827 out_err:
7828         cp = libbpf_strerror_r(-err, errmsg, sizeof(errmsg));
7829         pr_warn("failed to pin map: %s\n", cp);
7830         return err;
7831 }
7832
7833 int bpf_map__unpin(struct bpf_map *map, const char *path)
7834 {
7835         int err;
7836
7837         if (map == NULL) {
7838                 pr_warn("invalid map pointer\n");
7839                 return -EINVAL;
7840         }
7841
7842         if (map->pin_path) {
7843                 if (path && strcmp(path, map->pin_path)) {
7844                         pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
7845                                 bpf_map__name(map), map->pin_path, path);
7846                         return -EINVAL;
7847                 }
7848                 path = map->pin_path;
7849         } else if (!path) {
7850                 pr_warn("no path to unpin map '%s' from\n",
7851                         bpf_map__name(map));
7852                 return -EINVAL;
7853         }
7854
7855         err = check_path(path);
7856         if (err)
7857                 return err;
7858
7859         err = unlink(path);
7860         if (err != 0)
7861                 return -errno;
7862
7863         map->pinned = false;
7864         pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
7865
7866         return 0;
7867 }
7868
7869 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
7870 {
7871         char *new = NULL;
7872
7873         if (path) {
7874                 new = strdup(path);
7875                 if (!new)
7876                         return -errno;
7877         }
7878
7879         free(map->pin_path);
7880         map->pin_path = new;
7881         return 0;
7882 }
7883
7884 const char *bpf_map__get_pin_path(const struct bpf_map *map)
7885 {
7886         return map->pin_path;
7887 }
7888
7889 bool bpf_map__is_pinned(const struct bpf_map *map)
7890 {
7891         return map->pinned;
7892 }
7893
7894 static void sanitize_pin_path(char *s)
7895 {
7896         /* bpffs disallows periods in path names */
7897         while (*s) {
7898                 if (*s == '.')
7899                         *s = '_';
7900                 s++;
7901         }
7902 }
7903
7904 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
7905 {
7906         struct bpf_map *map;
7907         int err;
7908
7909         if (!obj)
7910                 return -ENOENT;
7911
7912         if (!obj->loaded) {
7913                 pr_warn("object not yet loaded; load it first\n");
7914                 return -ENOENT;
7915         }
7916
7917         bpf_object__for_each_map(map, obj) {
7918                 char *pin_path = NULL;
7919                 char buf[PATH_MAX];
7920
7921                 if (path) {
7922                         int len;
7923
7924                         len = snprintf(buf, PATH_MAX, "%s/%s", path,
7925                                        bpf_map__name(map));
7926                         if (len < 0) {
7927                                 err = -EINVAL;
7928                                 goto err_unpin_maps;
7929                         } else if (len >= PATH_MAX) {
7930                                 err = -ENAMETOOLONG;
7931                                 goto err_unpin_maps;
7932                         }
7933                         sanitize_pin_path(buf);
7934                         pin_path = buf;
7935                 } else if (!map->pin_path) {
7936                         continue;
7937                 }
7938
7939                 err = bpf_map__pin(map, pin_path);
7940                 if (err)
7941                         goto err_unpin_maps;
7942         }
7943
7944         return 0;
7945
7946 err_unpin_maps:
7947         while ((map = bpf_map__prev(map, obj))) {
7948                 if (!map->pin_path)
7949                         continue;
7950
7951                 bpf_map__unpin(map, NULL);
7952         }
7953
7954         return err;
7955 }
7956
7957 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
7958 {
7959         struct bpf_map *map;
7960         int err;
7961
7962         if (!obj)
7963                 return -ENOENT;
7964
7965         bpf_object__for_each_map(map, obj) {
7966                 char *pin_path = NULL;
7967                 char buf[PATH_MAX];
7968
7969                 if (path) {
7970                         int len;
7971
7972                         len = snprintf(buf, PATH_MAX, "%s/%s", path,
7973                                        bpf_map__name(map));
7974                         if (len < 0)
7975                                 return -EINVAL;
7976                         else if (len >= PATH_MAX)
7977                                 return -ENAMETOOLONG;
7978                         sanitize_pin_path(buf);
7979                         pin_path = buf;
7980                 } else if (!map->pin_path) {
7981                         continue;
7982                 }
7983
7984                 err = bpf_map__unpin(map, pin_path);
7985                 if (err)
7986                         return err;
7987         }
7988
7989         return 0;
7990 }
7991
7992 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
7993 {
7994         struct bpf_program *prog;
7995         int err;
7996
7997         if (!obj)
7998                 return -ENOENT;
7999
8000         if (!obj->loaded) {
8001                 pr_warn("object not yet loaded; load it first\n");
8002                 return -ENOENT;
8003         }
8004
8005         bpf_object__for_each_program(prog, obj) {
8006                 char buf[PATH_MAX];
8007                 int len;
8008
8009                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
8010                                prog->pin_name);
8011                 if (len < 0) {
8012                         err = -EINVAL;
8013                         goto err_unpin_programs;
8014                 } else if (len >= PATH_MAX) {
8015                         err = -ENAMETOOLONG;
8016                         goto err_unpin_programs;
8017                 }
8018
8019                 err = bpf_program__pin(prog, buf);
8020                 if (err)
8021                         goto err_unpin_programs;
8022         }
8023
8024         return 0;
8025
8026 err_unpin_programs:
8027         while ((prog = bpf_program__prev(prog, obj))) {
8028                 char buf[PATH_MAX];
8029                 int len;
8030
8031                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
8032                                prog->pin_name);
8033                 if (len < 0)
8034                         continue;
8035                 else if (len >= PATH_MAX)
8036                         continue;
8037
8038                 bpf_program__unpin(prog, buf);
8039         }
8040
8041         return err;
8042 }
8043
8044 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
8045 {
8046         struct bpf_program *prog;
8047         int err;
8048
8049         if (!obj)
8050                 return -ENOENT;
8051
8052         bpf_object__for_each_program(prog, obj) {
8053                 char buf[PATH_MAX];
8054                 int len;
8055
8056                 len = snprintf(buf, PATH_MAX, "%s/%s", path,
8057                                prog->pin_name);
8058                 if (len < 0)
8059                         return -EINVAL;
8060                 else if (len >= PATH_MAX)
8061                         return -ENAMETOOLONG;
8062
8063                 err = bpf_program__unpin(prog, buf);
8064                 if (err)
8065                         return err;
8066         }
8067
8068         return 0;
8069 }
8070
8071 int bpf_object__pin(struct bpf_object *obj, const char *path)
8072 {
8073         int err;
8074
8075         err = bpf_object__pin_maps(obj, path);
8076         if (err)
8077                 return err;
8078
8079         err = bpf_object__pin_programs(obj, path);
8080         if (err) {
8081                 bpf_object__unpin_maps(obj, path);
8082                 return err;
8083         }
8084
8085         return 0;
8086 }
8087
8088 static void bpf_map__destroy(struct bpf_map *map)
8089 {
8090         if (map->clear_priv)
8091                 map->clear_priv(map, map->priv);
8092         map->priv = NULL;
8093         map->clear_priv = NULL;
8094
8095         if (map->inner_map) {
8096                 bpf_map__destroy(map->inner_map);
8097                 zfree(&map->inner_map);
8098         }
8099
8100         zfree(&map->init_slots);
8101         map->init_slots_sz = 0;
8102
8103         if (map->mmaped) {
8104                 munmap(map->mmaped, bpf_map_mmap_sz(map));
8105                 map->mmaped = NULL;
8106         }
8107
8108         if (map->st_ops) {
8109                 zfree(&map->st_ops->data);
8110                 zfree(&map->st_ops->progs);
8111                 zfree(&map->st_ops->kern_func_off);
8112                 zfree(&map->st_ops);
8113         }
8114
8115         zfree(&map->name);
8116         zfree(&map->pin_path);
8117
8118         if (map->fd >= 0)
8119                 zclose(map->fd);
8120 }
8121
8122 void bpf_object__close(struct bpf_object *obj)
8123 {
8124         size_t i;
8125
8126         if (IS_ERR_OR_NULL(obj))
8127                 return;
8128
8129         if (obj->clear_priv)
8130                 obj->clear_priv(obj, obj->priv);
8131
8132         bpf_object__elf_finish(obj);
8133         bpf_object__unload(obj);
8134         btf__free(obj->btf);
8135         btf_ext__free(obj->btf_ext);
8136
8137         for (i = 0; i < obj->nr_maps; i++)
8138                 bpf_map__destroy(&obj->maps[i]);
8139
8140         zfree(&obj->kconfig);
8141         zfree(&obj->externs);
8142         obj->nr_extern = 0;
8143
8144         zfree(&obj->maps);
8145         obj->nr_maps = 0;
8146
8147         if (obj->programs && obj->nr_programs) {
8148                 for (i = 0; i < obj->nr_programs; i++)
8149                         bpf_program__exit(&obj->programs[i]);
8150         }
8151         zfree(&obj->programs);
8152
8153         list_del(&obj->list);
8154         free(obj);
8155 }
8156
8157 struct bpf_object *
8158 bpf_object__next(struct bpf_object *prev)
8159 {
8160         struct bpf_object *next;
8161
8162         if (!prev)
8163                 next = list_first_entry(&bpf_objects_list,
8164                                         struct bpf_object,
8165                                         list);
8166         else
8167                 next = list_next_entry(prev, list);
8168
8169         /* Empty list is noticed here so don't need checking on entry. */
8170         if (&next->list == &bpf_objects_list)
8171                 return NULL;
8172
8173         return next;
8174 }
8175
8176 const char *bpf_object__name(const struct bpf_object *obj)
8177 {
8178         return obj ? obj->name : ERR_PTR(-EINVAL);
8179 }
8180
8181 unsigned int bpf_object__kversion(const struct bpf_object *obj)
8182 {
8183         return obj ? obj->kern_version : 0;
8184 }
8185
8186 struct btf *bpf_object__btf(const struct bpf_object *obj)
8187 {
8188         return obj ? obj->btf : NULL;
8189 }
8190
8191 int bpf_object__btf_fd(const struct bpf_object *obj)
8192 {
8193         return obj->btf ? btf__fd(obj->btf) : -1;
8194 }
8195
8196 int bpf_object__set_priv(struct bpf_object *obj, void *priv,
8197                          bpf_object_clear_priv_t clear_priv)
8198 {
8199         if (obj->priv && obj->clear_priv)
8200                 obj->clear_priv(obj, obj->priv);
8201
8202         obj->priv = priv;
8203         obj->clear_priv = clear_priv;
8204         return 0;
8205 }
8206
8207 void *bpf_object__priv(const struct bpf_object *obj)
8208 {
8209         return obj ? obj->priv : ERR_PTR(-EINVAL);
8210 }
8211
8212 static struct bpf_program *
8213 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
8214                     bool forward)
8215 {
8216         size_t nr_programs = obj->nr_programs;
8217         ssize_t idx;
8218
8219         if (!nr_programs)
8220                 return NULL;
8221
8222         if (!p)
8223                 /* Iter from the beginning */
8224                 return forward ? &obj->programs[0] :
8225                         &obj->programs[nr_programs - 1];
8226
8227         if (p->obj != obj) {
8228                 pr_warn("error: program handler doesn't match object\n");
8229                 return NULL;
8230         }
8231
8232         idx = (p - obj->programs) + (forward ? 1 : -1);
8233         if (idx >= obj->nr_programs || idx < 0)
8234                 return NULL;
8235         return &obj->programs[idx];
8236 }
8237
8238 struct bpf_program *
8239 bpf_program__next(struct bpf_program *prev, const struct bpf_object *obj)
8240 {
8241         struct bpf_program *prog = prev;
8242
8243         do {
8244                 prog = __bpf_program__iter(prog, obj, true);
8245         } while (prog && prog_is_subprog(obj, prog));
8246
8247         return prog;
8248 }
8249
8250 struct bpf_program *
8251 bpf_program__prev(struct bpf_program *next, const struct bpf_object *obj)
8252 {
8253         struct bpf_program *prog = next;
8254
8255         do {
8256                 prog = __bpf_program__iter(prog, obj, false);
8257         } while (prog && prog_is_subprog(obj, prog));
8258
8259         return prog;
8260 }
8261
8262 int bpf_program__set_priv(struct bpf_program *prog, void *priv,
8263                           bpf_program_clear_priv_t clear_priv)
8264 {
8265         if (prog->priv && prog->clear_priv)
8266                 prog->clear_priv(prog, prog->priv);
8267
8268         prog->priv = priv;
8269         prog->clear_priv = clear_priv;
8270         return 0;
8271 }
8272
8273 void *bpf_program__priv(const struct bpf_program *prog)
8274 {
8275         return prog ? prog->priv : ERR_PTR(-EINVAL);
8276 }
8277
8278 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
8279 {
8280         prog->prog_ifindex = ifindex;
8281 }
8282
8283 const char *bpf_program__name(const struct bpf_program *prog)
8284 {
8285         return prog->name;
8286 }
8287
8288 const char *bpf_program__section_name(const struct bpf_program *prog)
8289 {
8290         return prog->sec_name;
8291 }
8292
8293 const char *bpf_program__title(const struct bpf_program *prog, bool needs_copy)
8294 {
8295         const char *title;
8296
8297         title = prog->sec_name;
8298         if (needs_copy) {
8299                 title = strdup(title);
8300                 if (!title) {
8301                         pr_warn("failed to strdup program title\n");
8302                         return ERR_PTR(-ENOMEM);
8303                 }
8304         }
8305
8306         return title;
8307 }
8308
8309 bool bpf_program__autoload(const struct bpf_program *prog)
8310 {
8311         return prog->load;
8312 }
8313
8314 int bpf_program__set_autoload(struct bpf_program *prog, bool autoload)
8315 {
8316         if (prog->obj->loaded)
8317                 return -EINVAL;
8318
8319         prog->load = autoload;
8320         return 0;
8321 }
8322
8323 int bpf_program__fd(const struct bpf_program *prog)
8324 {
8325         return bpf_program__nth_fd(prog, 0);
8326 }
8327
8328 size_t bpf_program__size(const struct bpf_program *prog)
8329 {
8330         return prog->insns_cnt * BPF_INSN_SZ;
8331 }
8332
8333 int bpf_program__set_prep(struct bpf_program *prog, int nr_instances,
8334                           bpf_program_prep_t prep)
8335 {
8336         int *instances_fds;
8337
8338         if (nr_instances <= 0 || !prep)
8339                 return -EINVAL;
8340
8341         if (prog->instances.nr > 0 || prog->instances.fds) {
8342                 pr_warn("Can't set pre-processor after loading\n");
8343                 return -EINVAL;
8344         }
8345
8346         instances_fds = malloc(sizeof(int) * nr_instances);
8347         if (!instances_fds) {
8348                 pr_warn("alloc memory failed for fds\n");
8349                 return -ENOMEM;
8350         }
8351
8352         /* fill all fd with -1 */
8353         memset(instances_fds, -1, sizeof(int) * nr_instances);
8354
8355         prog->instances.nr = nr_instances;
8356         prog->instances.fds = instances_fds;
8357         prog->preprocessor = prep;
8358         return 0;
8359 }
8360
8361 int bpf_program__nth_fd(const struct bpf_program *prog, int n)
8362 {
8363         int fd;
8364
8365         if (!prog)
8366                 return -EINVAL;
8367
8368         if (n >= prog->instances.nr || n < 0) {
8369                 pr_warn("Can't get the %dth fd from program %s: only %d instances\n",
8370                         n, prog->name, prog->instances.nr);
8371                 return -EINVAL;
8372         }
8373
8374         fd = prog->instances.fds[n];
8375         if (fd < 0) {
8376                 pr_warn("%dth instance of program '%s' is invalid\n",
8377                         n, prog->name);
8378                 return -ENOENT;
8379         }
8380
8381         return fd;
8382 }
8383
8384 enum bpf_prog_type bpf_program__get_type(struct bpf_program *prog)
8385 {
8386         return prog->type;
8387 }
8388
8389 void bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
8390 {
8391         prog->type = type;
8392 }
8393
8394 static bool bpf_program__is_type(const struct bpf_program *prog,
8395                                  enum bpf_prog_type type)
8396 {
8397         return prog ? (prog->type == type) : false;
8398 }
8399
8400 #define BPF_PROG_TYPE_FNS(NAME, TYPE)                           \
8401 int bpf_program__set_##NAME(struct bpf_program *prog)           \
8402 {                                                               \
8403         if (!prog)                                              \
8404                 return -EINVAL;                                 \
8405         bpf_program__set_type(prog, TYPE);                      \
8406         return 0;                                               \
8407 }                                                               \
8408                                                                 \
8409 bool bpf_program__is_##NAME(const struct bpf_program *prog)     \
8410 {                                                               \
8411         return bpf_program__is_type(prog, TYPE);                \
8412 }                                                               \
8413
8414 BPF_PROG_TYPE_FNS(socket_filter, BPF_PROG_TYPE_SOCKET_FILTER);
8415 BPF_PROG_TYPE_FNS(lsm, BPF_PROG_TYPE_LSM);
8416 BPF_PROG_TYPE_FNS(kprobe, BPF_PROG_TYPE_KPROBE);
8417 BPF_PROG_TYPE_FNS(sched_cls, BPF_PROG_TYPE_SCHED_CLS);
8418 BPF_PROG_TYPE_FNS(sched_act, BPF_PROG_TYPE_SCHED_ACT);
8419 BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
8420 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
8421 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
8422 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
8423 BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING);
8424 BPF_PROG_TYPE_FNS(struct_ops, BPF_PROG_TYPE_STRUCT_OPS);
8425 BPF_PROG_TYPE_FNS(extension, BPF_PROG_TYPE_EXT);
8426 BPF_PROG_TYPE_FNS(sk_lookup, BPF_PROG_TYPE_SK_LOOKUP);
8427
8428 enum bpf_attach_type
8429 bpf_program__get_expected_attach_type(struct bpf_program *prog)
8430 {
8431         return prog->expected_attach_type;
8432 }
8433
8434 void bpf_program__set_expected_attach_type(struct bpf_program *prog,
8435                                            enum bpf_attach_type type)
8436 {
8437         prog->expected_attach_type = type;
8438 }
8439
8440 #define BPF_PROG_SEC_IMPL(string, ptype, eatype, eatype_optional,           \
8441                           attachable, attach_btf)                           \
8442         {                                                                   \
8443                 .sec = string,                                              \
8444                 .len = sizeof(string) - 1,                                  \
8445                 .prog_type = ptype,                                         \
8446                 .expected_attach_type = eatype,                             \
8447                 .is_exp_attach_type_optional = eatype_optional,             \
8448                 .is_attachable = attachable,                                \
8449                 .is_attach_btf = attach_btf,                                \
8450         }
8451
8452 /* Programs that can NOT be attached. */
8453 #define BPF_PROG_SEC(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 0, 0)
8454
8455 /* Programs that can be attached. */
8456 #define BPF_APROG_SEC(string, ptype, atype) \
8457         BPF_PROG_SEC_IMPL(string, ptype, atype, true, 1, 0)
8458
8459 /* Programs that must specify expected attach type at load time. */
8460 #define BPF_EAPROG_SEC(string, ptype, eatype) \
8461         BPF_PROG_SEC_IMPL(string, ptype, eatype, false, 1, 0)
8462
8463 /* Programs that use BTF to identify attach point */
8464 #define BPF_PROG_BTF(string, ptype, eatype) \
8465         BPF_PROG_SEC_IMPL(string, ptype, eatype, false, 0, 1)
8466
8467 /* Programs that can be attached but attach type can't be identified by section
8468  * name. Kept for backward compatibility.
8469  */
8470 #define BPF_APROG_COMPAT(string, ptype) BPF_PROG_SEC(string, ptype)
8471
8472 #define SEC_DEF(sec_pfx, ptype, ...) {                                      \
8473         .sec = sec_pfx,                                                     \
8474         .len = sizeof(sec_pfx) - 1,                                         \
8475         .prog_type = BPF_PROG_TYPE_##ptype,                                 \
8476         __VA_ARGS__                                                         \
8477 }
8478
8479 static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
8480                                       struct bpf_program *prog);
8481 static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
8482                                   struct bpf_program *prog);
8483 static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
8484                                       struct bpf_program *prog);
8485 static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
8486                                      struct bpf_program *prog);
8487 static struct bpf_link *attach_lsm(const struct bpf_sec_def *sec,
8488                                    struct bpf_program *prog);
8489 static struct bpf_link *attach_iter(const struct bpf_sec_def *sec,
8490                                     struct bpf_program *prog);
8491
8492 static const struct bpf_sec_def section_defs[] = {
8493         BPF_PROG_SEC("socket",                  BPF_PROG_TYPE_SOCKET_FILTER),
8494         BPF_PROG_SEC("sk_reuseport",            BPF_PROG_TYPE_SK_REUSEPORT),
8495         SEC_DEF("kprobe/", KPROBE,
8496                 .attach_fn = attach_kprobe),
8497         BPF_PROG_SEC("uprobe/",                 BPF_PROG_TYPE_KPROBE),
8498         SEC_DEF("kretprobe/", KPROBE,
8499                 .attach_fn = attach_kprobe),
8500         BPF_PROG_SEC("uretprobe/",              BPF_PROG_TYPE_KPROBE),
8501         BPF_PROG_SEC("classifier",              BPF_PROG_TYPE_SCHED_CLS),
8502         BPF_PROG_SEC("action",                  BPF_PROG_TYPE_SCHED_ACT),
8503         SEC_DEF("tracepoint/", TRACEPOINT,
8504                 .attach_fn = attach_tp),
8505         SEC_DEF("tp/", TRACEPOINT,
8506                 .attach_fn = attach_tp),
8507         SEC_DEF("raw_tracepoint/", RAW_TRACEPOINT,
8508                 .attach_fn = attach_raw_tp),
8509         SEC_DEF("raw_tp/", RAW_TRACEPOINT,
8510                 .attach_fn = attach_raw_tp),
8511         SEC_DEF("tp_btf/", TRACING,
8512                 .expected_attach_type = BPF_TRACE_RAW_TP,
8513                 .is_attach_btf = true,
8514                 .attach_fn = attach_trace),
8515         SEC_DEF("fentry/", TRACING,
8516                 .expected_attach_type = BPF_TRACE_FENTRY,
8517                 .is_attach_btf = true,
8518                 .attach_fn = attach_trace),
8519         SEC_DEF("fmod_ret/", TRACING,
8520                 .expected_attach_type = BPF_MODIFY_RETURN,
8521                 .is_attach_btf = true,
8522                 .attach_fn = attach_trace),
8523         SEC_DEF("fexit/", TRACING,
8524                 .expected_attach_type = BPF_TRACE_FEXIT,
8525                 .is_attach_btf = true,
8526                 .attach_fn = attach_trace),
8527         SEC_DEF("fentry.s/", TRACING,
8528                 .expected_attach_type = BPF_TRACE_FENTRY,
8529                 .is_attach_btf = true,
8530                 .is_sleepable = true,
8531                 .attach_fn = attach_trace),
8532         SEC_DEF("fmod_ret.s/", TRACING,
8533                 .expected_attach_type = BPF_MODIFY_RETURN,
8534                 .is_attach_btf = true,
8535                 .is_sleepable = true,
8536                 .attach_fn = attach_trace),
8537         SEC_DEF("fexit.s/", TRACING,
8538                 .expected_attach_type = BPF_TRACE_FEXIT,
8539                 .is_attach_btf = true,
8540                 .is_sleepable = true,
8541                 .attach_fn = attach_trace),
8542         SEC_DEF("freplace/", EXT,
8543                 .is_attach_btf = true,
8544                 .attach_fn = attach_trace),
8545         SEC_DEF("lsm/", LSM,
8546                 .is_attach_btf = true,
8547                 .expected_attach_type = BPF_LSM_MAC,
8548                 .attach_fn = attach_lsm),
8549         SEC_DEF("lsm.s/", LSM,
8550                 .is_attach_btf = true,
8551                 .is_sleepable = true,
8552                 .expected_attach_type = BPF_LSM_MAC,
8553                 .attach_fn = attach_lsm),
8554         SEC_DEF("iter/", TRACING,
8555                 .expected_attach_type = BPF_TRACE_ITER,
8556                 .is_attach_btf = true,
8557                 .attach_fn = attach_iter),
8558         BPF_EAPROG_SEC("xdp_devmap/",           BPF_PROG_TYPE_XDP,
8559                                                 BPF_XDP_DEVMAP),
8560         BPF_EAPROG_SEC("xdp_cpumap/",           BPF_PROG_TYPE_XDP,
8561                                                 BPF_XDP_CPUMAP),
8562         BPF_APROG_SEC("xdp",                    BPF_PROG_TYPE_XDP,
8563                                                 BPF_XDP),
8564         BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
8565         BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
8566         BPF_PROG_SEC("lwt_out",                 BPF_PROG_TYPE_LWT_OUT),
8567         BPF_PROG_SEC("lwt_xmit",                BPF_PROG_TYPE_LWT_XMIT),
8568         BPF_PROG_SEC("lwt_seg6local",           BPF_PROG_TYPE_LWT_SEG6LOCAL),
8569         BPF_APROG_SEC("cgroup_skb/ingress",     BPF_PROG_TYPE_CGROUP_SKB,
8570                                                 BPF_CGROUP_INET_INGRESS),
8571         BPF_APROG_SEC("cgroup_skb/egress",      BPF_PROG_TYPE_CGROUP_SKB,
8572                                                 BPF_CGROUP_INET_EGRESS),
8573         BPF_APROG_COMPAT("cgroup/skb",          BPF_PROG_TYPE_CGROUP_SKB),
8574         BPF_EAPROG_SEC("cgroup/sock_create",    BPF_PROG_TYPE_CGROUP_SOCK,
8575                                                 BPF_CGROUP_INET_SOCK_CREATE),
8576         BPF_EAPROG_SEC("cgroup/sock_release",   BPF_PROG_TYPE_CGROUP_SOCK,
8577                                                 BPF_CGROUP_INET_SOCK_RELEASE),
8578         BPF_APROG_SEC("cgroup/sock",            BPF_PROG_TYPE_CGROUP_SOCK,
8579                                                 BPF_CGROUP_INET_SOCK_CREATE),
8580         BPF_EAPROG_SEC("cgroup/post_bind4",     BPF_PROG_TYPE_CGROUP_SOCK,
8581                                                 BPF_CGROUP_INET4_POST_BIND),
8582         BPF_EAPROG_SEC("cgroup/post_bind6",     BPF_PROG_TYPE_CGROUP_SOCK,
8583                                                 BPF_CGROUP_INET6_POST_BIND),
8584         BPF_APROG_SEC("cgroup/dev",             BPF_PROG_TYPE_CGROUP_DEVICE,
8585                                                 BPF_CGROUP_DEVICE),
8586         BPF_APROG_SEC("sockops",                BPF_PROG_TYPE_SOCK_OPS,
8587                                                 BPF_CGROUP_SOCK_OPS),
8588         BPF_APROG_SEC("sk_skb/stream_parser",   BPF_PROG_TYPE_SK_SKB,
8589                                                 BPF_SK_SKB_STREAM_PARSER),
8590         BPF_APROG_SEC("sk_skb/stream_verdict",  BPF_PROG_TYPE_SK_SKB,
8591                                                 BPF_SK_SKB_STREAM_VERDICT),
8592         BPF_APROG_COMPAT("sk_skb",              BPF_PROG_TYPE_SK_SKB),
8593         BPF_APROG_SEC("sk_msg",                 BPF_PROG_TYPE_SK_MSG,
8594                                                 BPF_SK_MSG_VERDICT),
8595         BPF_APROG_SEC("lirc_mode2",             BPF_PROG_TYPE_LIRC_MODE2,
8596                                                 BPF_LIRC_MODE2),
8597         BPF_APROG_SEC("flow_dissector",         BPF_PROG_TYPE_FLOW_DISSECTOR,
8598                                                 BPF_FLOW_DISSECTOR),
8599         BPF_EAPROG_SEC("cgroup/bind4",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8600                                                 BPF_CGROUP_INET4_BIND),
8601         BPF_EAPROG_SEC("cgroup/bind6",          BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8602                                                 BPF_CGROUP_INET6_BIND),
8603         BPF_EAPROG_SEC("cgroup/connect4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8604                                                 BPF_CGROUP_INET4_CONNECT),
8605         BPF_EAPROG_SEC("cgroup/connect6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8606                                                 BPF_CGROUP_INET6_CONNECT),
8607         BPF_EAPROG_SEC("cgroup/sendmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8608                                                 BPF_CGROUP_UDP4_SENDMSG),
8609         BPF_EAPROG_SEC("cgroup/sendmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8610                                                 BPF_CGROUP_UDP6_SENDMSG),
8611         BPF_EAPROG_SEC("cgroup/recvmsg4",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8612                                                 BPF_CGROUP_UDP4_RECVMSG),
8613         BPF_EAPROG_SEC("cgroup/recvmsg6",       BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8614                                                 BPF_CGROUP_UDP6_RECVMSG),
8615         BPF_EAPROG_SEC("cgroup/getpeername4",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8616                                                 BPF_CGROUP_INET4_GETPEERNAME),
8617         BPF_EAPROG_SEC("cgroup/getpeername6",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8618                                                 BPF_CGROUP_INET6_GETPEERNAME),
8619         BPF_EAPROG_SEC("cgroup/getsockname4",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8620                                                 BPF_CGROUP_INET4_GETSOCKNAME),
8621         BPF_EAPROG_SEC("cgroup/getsockname6",   BPF_PROG_TYPE_CGROUP_SOCK_ADDR,
8622                                                 BPF_CGROUP_INET6_GETSOCKNAME),
8623         BPF_EAPROG_SEC("cgroup/sysctl",         BPF_PROG_TYPE_CGROUP_SYSCTL,
8624                                                 BPF_CGROUP_SYSCTL),
8625         BPF_EAPROG_SEC("cgroup/getsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
8626                                                 BPF_CGROUP_GETSOCKOPT),
8627         BPF_EAPROG_SEC("cgroup/setsockopt",     BPF_PROG_TYPE_CGROUP_SOCKOPT,
8628                                                 BPF_CGROUP_SETSOCKOPT),
8629         BPF_PROG_SEC("struct_ops",              BPF_PROG_TYPE_STRUCT_OPS),
8630         BPF_EAPROG_SEC("sk_lookup/",            BPF_PROG_TYPE_SK_LOOKUP,
8631                                                 BPF_SK_LOOKUP),
8632 };
8633
8634 #undef BPF_PROG_SEC_IMPL
8635 #undef BPF_PROG_SEC
8636 #undef BPF_APROG_SEC
8637 #undef BPF_EAPROG_SEC
8638 #undef BPF_APROG_COMPAT
8639 #undef SEC_DEF
8640
8641 #define MAX_TYPE_NAME_SIZE 32
8642
8643 static const struct bpf_sec_def *find_sec_def(const char *sec_name)
8644 {
8645         int i, n = ARRAY_SIZE(section_defs);
8646
8647         for (i = 0; i < n; i++) {
8648                 if (strncmp(sec_name,
8649                             section_defs[i].sec, section_defs[i].len))
8650                         continue;
8651                 return &section_defs[i];
8652         }
8653         return NULL;
8654 }
8655
8656 static char *libbpf_get_type_names(bool attach_type)
8657 {
8658         int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
8659         char *buf;
8660
8661         buf = malloc(len);
8662         if (!buf)
8663                 return NULL;
8664
8665         buf[0] = '\0';
8666         /* Forge string buf with all available names */
8667         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
8668                 if (attach_type && !section_defs[i].is_attachable)
8669                         continue;
8670
8671                 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
8672                         free(buf);
8673                         return NULL;
8674                 }
8675                 strcat(buf, " ");
8676                 strcat(buf, section_defs[i].sec);
8677         }
8678
8679         return buf;
8680 }
8681
8682 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
8683                              enum bpf_attach_type *expected_attach_type)
8684 {
8685         const struct bpf_sec_def *sec_def;
8686         char *type_names;
8687
8688         if (!name)
8689                 return -EINVAL;
8690
8691         sec_def = find_sec_def(name);
8692         if (sec_def) {
8693                 *prog_type = sec_def->prog_type;
8694                 *expected_attach_type = sec_def->expected_attach_type;
8695                 return 0;
8696         }
8697
8698         pr_debug("failed to guess program type from ELF section '%s'\n", name);
8699         type_names = libbpf_get_type_names(false);
8700         if (type_names != NULL) {
8701                 pr_debug("supported section(type) names are:%s\n", type_names);
8702                 free(type_names);
8703         }
8704
8705         return -ESRCH;
8706 }
8707
8708 static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
8709                                                      size_t offset)
8710 {
8711         struct bpf_map *map;
8712         size_t i;
8713
8714         for (i = 0; i < obj->nr_maps; i++) {
8715                 map = &obj->maps[i];
8716                 if (!bpf_map__is_struct_ops(map))
8717                         continue;
8718                 if (map->sec_offset <= offset &&
8719                     offset - map->sec_offset < map->def.value_size)
8720                         return map;
8721         }
8722
8723         return NULL;
8724 }
8725
8726 /* Collect the reloc from ELF and populate the st_ops->progs[] */
8727 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
8728                                             GElf_Shdr *shdr, Elf_Data *data)
8729 {
8730         const struct btf_member *member;
8731         struct bpf_struct_ops *st_ops;
8732         struct bpf_program *prog;
8733         unsigned int shdr_idx;
8734         const struct btf *btf;
8735         struct bpf_map *map;
8736         Elf_Data *symbols;
8737         unsigned int moff, insn_idx;
8738         const char *name;
8739         __u32 member_idx;
8740         GElf_Sym sym;
8741         GElf_Rel rel;
8742         int i, nrels;
8743
8744         symbols = obj->efile.symbols;
8745         btf = obj->btf;
8746         nrels = shdr->sh_size / shdr->sh_entsize;
8747         for (i = 0; i < nrels; i++) {
8748                 if (!gelf_getrel(data, i, &rel)) {
8749                         pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
8750                         return -LIBBPF_ERRNO__FORMAT;
8751                 }
8752
8753                 if (!gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &sym)) {
8754                         pr_warn("struct_ops reloc: symbol %zx not found\n",
8755                                 (size_t)GELF_R_SYM(rel.r_info));
8756                         return -LIBBPF_ERRNO__FORMAT;
8757                 }
8758
8759                 name = elf_sym_str(obj, sym.st_name) ?: "<?>";
8760                 map = find_struct_ops_map_by_offset(obj, rel.r_offset);
8761                 if (!map) {
8762                         pr_warn("struct_ops reloc: cannot find map at rel.r_offset %zu\n",
8763                                 (size_t)rel.r_offset);
8764                         return -EINVAL;
8765                 }
8766
8767                 moff = rel.r_offset - map->sec_offset;
8768                 shdr_idx = sym.st_shndx;
8769                 st_ops = map->st_ops;
8770                 pr_debug("struct_ops reloc %s: for %lld value %lld shdr_idx %u rel.r_offset %zu map->sec_offset %zu name %d (\'%s\')\n",
8771                          map->name,
8772                          (long long)(rel.r_info >> 32),
8773                          (long long)sym.st_value,
8774                          shdr_idx, (size_t)rel.r_offset,
8775                          map->sec_offset, sym.st_name, name);
8776
8777                 if (shdr_idx >= SHN_LORESERVE) {
8778                         pr_warn("struct_ops reloc %s: rel.r_offset %zu shdr_idx %u unsupported non-static function\n",
8779                                 map->name, (size_t)rel.r_offset, shdr_idx);
8780                         return -LIBBPF_ERRNO__RELOC;
8781                 }
8782                 if (sym.st_value % BPF_INSN_SZ) {
8783                         pr_warn("struct_ops reloc %s: invalid target program offset %llu\n",
8784                                 map->name, (unsigned long long)sym.st_value);
8785                         return -LIBBPF_ERRNO__FORMAT;
8786                 }
8787                 insn_idx = sym.st_value / BPF_INSN_SZ;
8788
8789                 member = find_member_by_offset(st_ops->type, moff * 8);
8790                 if (!member) {
8791                         pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
8792                                 map->name, moff);
8793                         return -EINVAL;
8794                 }
8795                 member_idx = member - btf_members(st_ops->type);
8796                 name = btf__name_by_offset(btf, member->name_off);
8797
8798                 if (!resolve_func_ptr(btf, member->type, NULL)) {
8799                         pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
8800                                 map->name, name);
8801                         return -EINVAL;
8802                 }
8803
8804                 prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx);
8805                 if (!prog) {
8806                         pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
8807                                 map->name, shdr_idx, name);
8808                         return -EINVAL;
8809                 }
8810
8811                 if (prog->type == BPF_PROG_TYPE_UNSPEC) {
8812                         const struct bpf_sec_def *sec_def;
8813
8814                         sec_def = find_sec_def(prog->sec_name);
8815                         if (sec_def &&
8816                             sec_def->prog_type != BPF_PROG_TYPE_STRUCT_OPS) {
8817                                 /* for pr_warn */
8818                                 prog->type = sec_def->prog_type;
8819                                 goto invalid_prog;
8820                         }
8821
8822                         prog->type = BPF_PROG_TYPE_STRUCT_OPS;
8823                         prog->attach_btf_id = st_ops->type_id;
8824                         prog->expected_attach_type = member_idx;
8825                 } else if (prog->type != BPF_PROG_TYPE_STRUCT_OPS ||
8826                            prog->attach_btf_id != st_ops->type_id ||
8827                            prog->expected_attach_type != member_idx) {
8828                         goto invalid_prog;
8829                 }
8830                 st_ops->progs[member_idx] = prog;
8831         }
8832
8833         return 0;
8834
8835 invalid_prog:
8836         pr_warn("struct_ops reloc %s: cannot use prog %s in sec %s with type %u attach_btf_id %u expected_attach_type %u for func ptr %s\n",
8837                 map->name, prog->name, prog->sec_name, prog->type,
8838                 prog->attach_btf_id, prog->expected_attach_type, name);
8839         return -EINVAL;
8840 }
8841
8842 #define BTF_TRACE_PREFIX "btf_trace_"
8843 #define BTF_LSM_PREFIX "bpf_lsm_"
8844 #define BTF_ITER_PREFIX "bpf_iter_"
8845 #define BTF_MAX_NAME_SIZE 128
8846
8847 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
8848                                    const char *name, __u32 kind)
8849 {
8850         char btf_type_name[BTF_MAX_NAME_SIZE];
8851         int ret;
8852
8853         ret = snprintf(btf_type_name, sizeof(btf_type_name),
8854                        "%s%s", prefix, name);
8855         /* snprintf returns the number of characters written excluding the
8856          * the terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
8857          * indicates truncation.
8858          */
8859         if (ret < 0 || ret >= sizeof(btf_type_name))
8860                 return -ENAMETOOLONG;
8861         return btf__find_by_name_kind(btf, btf_type_name, kind);
8862 }
8863
8864 static inline int find_attach_btf_id(struct btf *btf, const char *name,
8865                                      enum bpf_attach_type attach_type)
8866 {
8867         int err;
8868
8869         if (attach_type == BPF_TRACE_RAW_TP)
8870                 err = find_btf_by_prefix_kind(btf, BTF_TRACE_PREFIX, name,
8871                                               BTF_KIND_TYPEDEF);
8872         else if (attach_type == BPF_LSM_MAC)
8873                 err = find_btf_by_prefix_kind(btf, BTF_LSM_PREFIX, name,
8874                                               BTF_KIND_FUNC);
8875         else if (attach_type == BPF_TRACE_ITER)
8876                 err = find_btf_by_prefix_kind(btf, BTF_ITER_PREFIX, name,
8877                                               BTF_KIND_FUNC);
8878         else
8879                 err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
8880
8881         return err;
8882 }
8883
8884 int libbpf_find_vmlinux_btf_id(const char *name,
8885                                enum bpf_attach_type attach_type)
8886 {
8887         struct btf *btf;
8888         int err;
8889
8890         btf = libbpf_find_kernel_btf();
8891         if (IS_ERR(btf)) {
8892                 pr_warn("vmlinux BTF is not found\n");
8893                 return -EINVAL;
8894         }
8895
8896         err = find_attach_btf_id(btf, name, attach_type);
8897         if (err <= 0)
8898                 pr_warn("%s is not found in vmlinux BTF\n", name);
8899
8900         btf__free(btf);
8901         return err;
8902 }
8903
8904 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd)
8905 {
8906         struct bpf_prog_info_linear *info_linear;
8907         struct bpf_prog_info *info;
8908         struct btf *btf = NULL;
8909         int err = -EINVAL;
8910
8911         info_linear = bpf_program__get_prog_info_linear(attach_prog_fd, 0);
8912         if (IS_ERR_OR_NULL(info_linear)) {
8913                 pr_warn("failed get_prog_info_linear for FD %d\n",
8914                         attach_prog_fd);
8915                 return -EINVAL;
8916         }
8917         info = &info_linear->info;
8918         if (!info->btf_id) {
8919                 pr_warn("The target program doesn't have BTF\n");
8920                 goto out;
8921         }
8922         if (btf__get_from_id(info->btf_id, &btf)) {
8923                 pr_warn("Failed to get BTF of the program\n");
8924                 goto out;
8925         }
8926         err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
8927         btf__free(btf);
8928         if (err <= 0) {
8929                 pr_warn("%s is not found in prog's BTF\n", name);
8930                 goto out;
8931         }
8932 out:
8933         free(info_linear);
8934         return err;
8935 }
8936
8937 static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name,
8938                               enum bpf_attach_type attach_type,
8939                               int *btf_obj_fd, int *btf_type_id)
8940 {
8941         int ret, i;
8942
8943         ret = find_attach_btf_id(obj->btf_vmlinux, attach_name, attach_type);
8944         if (ret > 0) {
8945                 *btf_obj_fd = 0; /* vmlinux BTF */
8946                 *btf_type_id = ret;
8947                 return 0;
8948         }
8949         if (ret != -ENOENT)
8950                 return ret;
8951
8952         ret = load_module_btfs(obj);
8953         if (ret)
8954                 return ret;
8955
8956         for (i = 0; i < obj->btf_module_cnt; i++) {
8957                 const struct module_btf *mod = &obj->btf_modules[i];
8958
8959                 ret = find_attach_btf_id(mod->btf, attach_name, attach_type);
8960                 if (ret > 0) {
8961                         *btf_obj_fd = mod->fd;
8962                         *btf_type_id = ret;
8963                         return 0;
8964                 }
8965                 if (ret == -ENOENT)
8966                         continue;
8967
8968                 return ret;
8969         }
8970
8971         return -ESRCH;
8972 }
8973
8974 static int libbpf_find_attach_btf_id(struct bpf_program *prog, int *btf_obj_fd, int *btf_type_id)
8975 {
8976         enum bpf_attach_type attach_type = prog->expected_attach_type;
8977         __u32 attach_prog_fd = prog->attach_prog_fd;
8978         const char *name = prog->sec_name, *attach_name;
8979         const struct bpf_sec_def *sec = NULL;
8980         int i, err;
8981
8982         if (!name)
8983                 return -EINVAL;
8984
8985         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
8986                 if (!section_defs[i].is_attach_btf)
8987                         continue;
8988                 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
8989                         continue;
8990
8991                 sec = &section_defs[i];
8992                 break;
8993         }
8994
8995         if (!sec) {
8996                 pr_warn("failed to identify BTF ID based on ELF section name '%s'\n", name);
8997                 return -ESRCH;
8998         }
8999         attach_name = name + sec->len;
9000
9001         /* BPF program's BTF ID */
9002         if (attach_prog_fd) {
9003                 err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd);
9004                 if (err < 0) {
9005                         pr_warn("failed to find BPF program (FD %d) BTF ID for '%s': %d\n",
9006                                  attach_prog_fd, attach_name, err);
9007                         return err;
9008                 }
9009                 *btf_obj_fd = 0;
9010                 *btf_type_id = err;
9011                 return 0;
9012         }
9013
9014         /* kernel/module BTF ID */
9015         err = find_kernel_btf_id(prog->obj, attach_name, attach_type, btf_obj_fd, btf_type_id);
9016         if (err) {
9017                 pr_warn("failed to find kernel BTF type ID of '%s': %d\n", attach_name, err);
9018                 return err;
9019         }
9020         return 0;
9021 }
9022
9023 int libbpf_attach_type_by_name(const char *name,
9024                                enum bpf_attach_type *attach_type)
9025 {
9026         char *type_names;
9027         int i;
9028
9029         if (!name)
9030                 return -EINVAL;
9031
9032         for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
9033                 if (strncmp(name, section_defs[i].sec, section_defs[i].len))
9034                         continue;
9035                 if (!section_defs[i].is_attachable)
9036                         return -EINVAL;
9037                 *attach_type = section_defs[i].expected_attach_type;
9038                 return 0;
9039         }
9040         pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
9041         type_names = libbpf_get_type_names(true);
9042         if (type_names != NULL) {
9043                 pr_debug("attachable section(type) names are:%s\n", type_names);
9044                 free(type_names);
9045         }
9046
9047         return -EINVAL;
9048 }
9049
9050 int bpf_map__fd(const struct bpf_map *map)
9051 {
9052         return map ? map->fd : -EINVAL;
9053 }
9054
9055 const struct bpf_map_def *bpf_map__def(const struct bpf_map *map)
9056 {
9057         return map ? &map->def : ERR_PTR(-EINVAL);
9058 }
9059
9060 const char *bpf_map__name(const struct bpf_map *map)
9061 {
9062         return map ? map->name : NULL;
9063 }
9064
9065 enum bpf_map_type bpf_map__type(const struct bpf_map *map)
9066 {
9067         return map->def.type;
9068 }
9069
9070 int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type)
9071 {
9072         if (map->fd >= 0)
9073                 return -EBUSY;
9074         map->def.type = type;
9075         return 0;
9076 }
9077
9078 __u32 bpf_map__map_flags(const struct bpf_map *map)
9079 {
9080         return map->def.map_flags;
9081 }
9082
9083 int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags)
9084 {
9085         if (map->fd >= 0)
9086                 return -EBUSY;
9087         map->def.map_flags = flags;
9088         return 0;
9089 }
9090
9091 __u32 bpf_map__numa_node(const struct bpf_map *map)
9092 {
9093         return map->numa_node;
9094 }
9095
9096 int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node)
9097 {
9098         if (map->fd >= 0)
9099                 return -EBUSY;
9100         map->numa_node = numa_node;
9101         return 0;
9102 }
9103
9104 __u32 bpf_map__key_size(const struct bpf_map *map)
9105 {
9106         return map->def.key_size;
9107 }
9108
9109 int bpf_map__set_key_size(struct bpf_map *map, __u32 size)
9110 {
9111         if (map->fd >= 0)
9112                 return -EBUSY;
9113         map->def.key_size = size;
9114         return 0;
9115 }
9116
9117 __u32 bpf_map__value_size(const struct bpf_map *map)
9118 {
9119         return map->def.value_size;
9120 }
9121
9122 int bpf_map__set_value_size(struct bpf_map *map, __u32 size)
9123 {
9124         if (map->fd >= 0)
9125                 return -EBUSY;
9126         map->def.value_size = size;
9127         return 0;
9128 }
9129
9130 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
9131 {
9132         return map ? map->btf_key_type_id : 0;
9133 }
9134
9135 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
9136 {
9137         return map ? map->btf_value_type_id : 0;
9138 }
9139
9140 int bpf_map__set_priv(struct bpf_map *map, void *priv,
9141                      bpf_map_clear_priv_t clear_priv)
9142 {
9143         if (!map)
9144                 return -EINVAL;
9145
9146         if (map->priv) {
9147                 if (map->clear_priv)
9148                         map->clear_priv(map, map->priv);
9149         }
9150
9151         map->priv = priv;
9152         map->clear_priv = clear_priv;
9153         return 0;
9154 }
9155
9156 void *bpf_map__priv(const struct bpf_map *map)
9157 {
9158         return map ? map->priv : ERR_PTR(-EINVAL);
9159 }
9160
9161 int bpf_map__set_initial_value(struct bpf_map *map,
9162                                const void *data, size_t size)
9163 {
9164         if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG ||
9165             size != map->def.value_size || map->fd >= 0)
9166                 return -EINVAL;
9167
9168         memcpy(map->mmaped, data, size);
9169         return 0;
9170 }
9171
9172 bool bpf_map__is_offload_neutral(const struct bpf_map *map)
9173 {
9174         return map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY;
9175 }
9176
9177 bool bpf_map__is_internal(const struct bpf_map *map)
9178 {
9179         return map->libbpf_type != LIBBPF_MAP_UNSPEC;
9180 }
9181
9182 __u32 bpf_map__ifindex(const struct bpf_map *map)
9183 {
9184         return map->map_ifindex;
9185 }
9186
9187 int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
9188 {
9189         if (map->fd >= 0)
9190                 return -EBUSY;
9191         map->map_ifindex = ifindex;
9192         return 0;
9193 }
9194
9195 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
9196 {
9197         if (!bpf_map_type__is_map_in_map(map->def.type)) {
9198                 pr_warn("error: unsupported map type\n");
9199                 return -EINVAL;
9200         }
9201         if (map->inner_map_fd != -1) {
9202                 pr_warn("error: inner_map_fd already specified\n");
9203                 return -EINVAL;
9204         }
9205         map->inner_map_fd = fd;
9206         return 0;
9207 }
9208
9209 static struct bpf_map *
9210 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
9211 {
9212         ssize_t idx;
9213         struct bpf_map *s, *e;
9214
9215         if (!obj || !obj->maps)
9216                 return NULL;
9217
9218         s = obj->maps;
9219         e = obj->maps + obj->nr_maps;
9220
9221         if ((m < s) || (m >= e)) {
9222                 pr_warn("error in %s: map handler doesn't belong to object\n",
9223                          __func__);
9224                 return NULL;
9225         }
9226
9227         idx = (m - obj->maps) + i;
9228         if (idx >= obj->nr_maps || idx < 0)
9229                 return NULL;
9230         return &obj->maps[idx];
9231 }
9232
9233 struct bpf_map *
9234 bpf_map__next(const struct bpf_map *prev, const struct bpf_object *obj)
9235 {
9236         if (prev == NULL)
9237                 return obj->maps;
9238
9239         return __bpf_map__iter(prev, obj, 1);
9240 }
9241
9242 struct bpf_map *
9243 bpf_map__prev(const struct bpf_map *next, const struct bpf_object *obj)
9244 {
9245         if (next == NULL) {
9246                 if (!obj->nr_maps)
9247                         return NULL;
9248                 return obj->maps + obj->nr_maps - 1;
9249         }
9250
9251         return __bpf_map__iter(next, obj, -1);
9252 }
9253
9254 struct bpf_map *
9255 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
9256 {
9257         struct bpf_map *pos;
9258
9259         bpf_object__for_each_map(pos, obj) {
9260                 if (pos->name && !strcmp(pos->name, name))
9261                         return pos;
9262         }
9263         return NULL;
9264 }
9265
9266 int
9267 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
9268 {
9269         return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
9270 }
9271
9272 struct bpf_map *
9273 bpf_object__find_map_by_offset(struct bpf_object *obj, size_t offset)
9274 {
9275         return ERR_PTR(-ENOTSUP);
9276 }
9277
9278 long libbpf_get_error(const void *ptr)
9279 {
9280         return PTR_ERR_OR_ZERO(ptr);
9281 }
9282
9283 int bpf_prog_load(const char *file, enum bpf_prog_type type,
9284                   struct bpf_object **pobj, int *prog_fd)
9285 {
9286         struct bpf_prog_load_attr attr;
9287
9288         memset(&attr, 0, sizeof(struct bpf_prog_load_attr));
9289         attr.file = file;
9290         attr.prog_type = type;
9291         attr.expected_attach_type = 0;
9292
9293         return bpf_prog_load_xattr(&attr, pobj, prog_fd);
9294 }
9295
9296 int bpf_prog_load_xattr(const struct bpf_prog_load_attr *attr,
9297                         struct bpf_object **pobj, int *prog_fd)
9298 {
9299         struct bpf_object_open_attr open_attr = {};
9300         struct bpf_program *prog, *first_prog = NULL;
9301         struct bpf_object *obj;
9302         struct bpf_map *map;
9303         int err;
9304
9305         if (!attr)
9306                 return -EINVAL;
9307         if (!attr->file)
9308                 return -EINVAL;
9309
9310         open_attr.file = attr->file;
9311         open_attr.prog_type = attr->prog_type;
9312
9313         obj = bpf_object__open_xattr(&open_attr);
9314         if (IS_ERR_OR_NULL(obj))
9315                 return -ENOENT;
9316
9317         bpf_object__for_each_program(prog, obj) {
9318                 enum bpf_attach_type attach_type = attr->expected_attach_type;
9319                 /*
9320                  * to preserve backwards compatibility, bpf_prog_load treats
9321                  * attr->prog_type, if specified, as an override to whatever
9322                  * bpf_object__open guessed
9323                  */
9324                 if (attr->prog_type != BPF_PROG_TYPE_UNSPEC) {
9325                         bpf_program__set_type(prog, attr->prog_type);
9326                         bpf_program__set_expected_attach_type(prog,
9327                                                               attach_type);
9328                 }
9329                 if (bpf_program__get_type(prog) == BPF_PROG_TYPE_UNSPEC) {
9330                         /*
9331                          * we haven't guessed from section name and user
9332                          * didn't provide a fallback type, too bad...
9333                          */
9334                         bpf_object__close(obj);
9335                         return -EINVAL;
9336                 }
9337
9338                 prog->prog_ifindex = attr->ifindex;
9339                 prog->log_level = attr->log_level;
9340                 prog->prog_flags |= attr->prog_flags;
9341                 if (!first_prog)
9342                         first_prog = prog;
9343         }
9344
9345         bpf_object__for_each_map(map, obj) {
9346                 if (!bpf_map__is_offload_neutral(map))
9347                         map->map_ifindex = attr->ifindex;
9348         }
9349
9350         if (!first_prog) {
9351                 pr_warn("object file doesn't contain bpf program\n");
9352                 bpf_object__close(obj);
9353                 return -ENOENT;
9354         }
9355
9356         err = bpf_object__load(obj);
9357         if (err) {
9358                 bpf_object__close(obj);
9359                 return err;
9360         }
9361
9362         *pobj = obj;
9363         *prog_fd = bpf_program__fd(first_prog);
9364         return 0;
9365 }
9366
9367 struct bpf_link {
9368         int (*detach)(struct bpf_link *link);
9369         int (*destroy)(struct bpf_link *link);
9370         char *pin_path;         /* NULL, if not pinned */
9371         int fd;                 /* hook FD, -1 if not applicable */
9372         bool disconnected;
9373 };
9374
9375 /* Replace link's underlying BPF program with the new one */
9376 int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog)
9377 {
9378         return bpf_link_update(bpf_link__fd(link), bpf_program__fd(prog), NULL);
9379 }
9380
9381 /* Release "ownership" of underlying BPF resource (typically, BPF program
9382  * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
9383  * link, when destructed through bpf_link__destroy() call won't attempt to
9384  * detach/unregisted that BPF resource. This is useful in situations where,
9385  * say, attached BPF program has to outlive userspace program that attached it
9386  * in the system. Depending on type of BPF program, though, there might be
9387  * additional steps (like pinning BPF program in BPF FS) necessary to ensure
9388  * exit of userspace program doesn't trigger automatic detachment and clean up
9389  * inside the kernel.
9390  */
9391 void bpf_link__disconnect(struct bpf_link *link)
9392 {
9393         link->disconnected = true;
9394 }
9395
9396 int bpf_link__destroy(struct bpf_link *link)
9397 {
9398         int err = 0;
9399
9400         if (IS_ERR_OR_NULL(link))
9401                 return 0;
9402
9403         if (!link->disconnected && link->detach)
9404                 err = link->detach(link);
9405         if (link->destroy)
9406                 link->destroy(link);
9407         if (link->pin_path)
9408                 free(link->pin_path);
9409         free(link);
9410
9411         return err;
9412 }
9413
9414 int bpf_link__fd(const struct bpf_link *link)
9415 {
9416         return link->fd;
9417 }
9418
9419 const char *bpf_link__pin_path(const struct bpf_link *link)
9420 {
9421         return link->pin_path;
9422 }
9423
9424 static int bpf_link__detach_fd(struct bpf_link *link)
9425 {
9426         return close(link->fd);
9427 }
9428
9429 struct bpf_link *bpf_link__open(const char *path)
9430 {
9431         struct bpf_link *link;
9432         int fd;
9433
9434         fd = bpf_obj_get(path);
9435         if (fd < 0) {
9436                 fd = -errno;
9437                 pr_warn("failed to open link at %s: %d\n", path, fd);
9438                 return ERR_PTR(fd);
9439         }
9440
9441         link = calloc(1, sizeof(*link));
9442         if (!link) {
9443                 close(fd);
9444                 return ERR_PTR(-ENOMEM);
9445         }
9446         link->detach = &bpf_link__detach_fd;
9447         link->fd = fd;
9448
9449         link->pin_path = strdup(path);
9450         if (!link->pin_path) {
9451                 bpf_link__destroy(link);
9452                 return ERR_PTR(-ENOMEM);
9453         }
9454
9455         return link;
9456 }
9457
9458 int bpf_link__detach(struct bpf_link *link)
9459 {
9460         return bpf_link_detach(link->fd) ? -errno : 0;
9461 }
9462
9463 int bpf_link__pin(struct bpf_link *link, const char *path)
9464 {
9465         int err;
9466
9467         if (link->pin_path)
9468                 return -EBUSY;
9469         err = make_parent_dir(path);
9470         if (err)
9471                 return err;
9472         err = check_path(path);
9473         if (err)
9474                 return err;
9475
9476         link->pin_path = strdup(path);
9477         if (!link->pin_path)
9478                 return -ENOMEM;
9479
9480         if (bpf_obj_pin(link->fd, link->pin_path)) {
9481                 err = -errno;
9482                 zfree(&link->pin_path);
9483                 return err;
9484         }
9485
9486         pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path);
9487         return 0;
9488 }
9489
9490 int bpf_link__unpin(struct bpf_link *link)
9491 {
9492         int err;
9493
9494         if (!link->pin_path)
9495                 return -EINVAL;
9496
9497         err = unlink(link->pin_path);
9498         if (err != 0)
9499                 return -errno;
9500
9501         pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path);
9502         zfree(&link->pin_path);
9503         return 0;
9504 }
9505
9506 static int bpf_link__detach_perf_event(struct bpf_link *link)
9507 {
9508         int err;
9509
9510         err = ioctl(link->fd, PERF_EVENT_IOC_DISABLE, 0);
9511         if (err)
9512                 err = -errno;
9513
9514         close(link->fd);
9515         return err;
9516 }
9517
9518 struct bpf_link *bpf_program__attach_perf_event(struct bpf_program *prog,
9519                                                 int pfd)
9520 {
9521         char errmsg[STRERR_BUFSIZE];
9522         struct bpf_link *link;
9523         int prog_fd, err;
9524
9525         if (pfd < 0) {
9526                 pr_warn("prog '%s': invalid perf event FD %d\n",
9527                         prog->name, pfd);
9528                 return ERR_PTR(-EINVAL);
9529         }
9530         prog_fd = bpf_program__fd(prog);
9531         if (prog_fd < 0) {
9532                 pr_warn("prog '%s': can't attach BPF program w/o FD (did you load it?)\n",
9533                         prog->name);
9534                 return ERR_PTR(-EINVAL);
9535         }
9536
9537         link = calloc(1, sizeof(*link));
9538         if (!link)
9539                 return ERR_PTR(-ENOMEM);
9540         link->detach = &bpf_link__detach_perf_event;
9541         link->fd = pfd;
9542
9543         if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
9544                 err = -errno;
9545                 free(link);
9546                 pr_warn("prog '%s': failed to attach to pfd %d: %s\n",
9547                         prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9548                 if (err == -EPROTO)
9549                         pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n",
9550                                 prog->name, pfd);
9551                 return ERR_PTR(err);
9552         }
9553         if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
9554                 err = -errno;
9555                 free(link);
9556                 pr_warn("prog '%s': failed to enable pfd %d: %s\n",
9557                         prog->name, pfd, libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9558                 return ERR_PTR(err);
9559         }
9560         return link;
9561 }
9562
9563 /*
9564  * this function is expected to parse integer in the range of [0, 2^31-1] from
9565  * given file using scanf format string fmt. If actual parsed value is
9566  * negative, the result might be indistinguishable from error
9567  */
9568 static int parse_uint_from_file(const char *file, const char *fmt)
9569 {
9570         char buf[STRERR_BUFSIZE];
9571         int err, ret;
9572         FILE *f;
9573
9574         f = fopen(file, "r");
9575         if (!f) {
9576                 err = -errno;
9577                 pr_debug("failed to open '%s': %s\n", file,
9578                          libbpf_strerror_r(err, buf, sizeof(buf)));
9579                 return err;
9580         }
9581         err = fscanf(f, fmt, &ret);
9582         if (err != 1) {
9583                 err = err == EOF ? -EIO : -errno;
9584                 pr_debug("failed to parse '%s': %s\n", file,
9585                         libbpf_strerror_r(err, buf, sizeof(buf)));
9586                 fclose(f);
9587                 return err;
9588         }
9589         fclose(f);
9590         return ret;
9591 }
9592
9593 static int determine_kprobe_perf_type(void)
9594 {
9595         const char *file = "/sys/bus/event_source/devices/kprobe/type";
9596
9597         return parse_uint_from_file(file, "%d\n");
9598 }
9599
9600 static int determine_uprobe_perf_type(void)
9601 {
9602         const char *file = "/sys/bus/event_source/devices/uprobe/type";
9603
9604         return parse_uint_from_file(file, "%d\n");
9605 }
9606
9607 static int determine_kprobe_retprobe_bit(void)
9608 {
9609         const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
9610
9611         return parse_uint_from_file(file, "config:%d\n");
9612 }
9613
9614 static int determine_uprobe_retprobe_bit(void)
9615 {
9616         const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
9617
9618         return parse_uint_from_file(file, "config:%d\n");
9619 }
9620
9621 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
9622                                  uint64_t offset, int pid)
9623 {
9624         struct perf_event_attr attr = {};
9625         char errmsg[STRERR_BUFSIZE];
9626         int type, pfd, err;
9627
9628         type = uprobe ? determine_uprobe_perf_type()
9629                       : determine_kprobe_perf_type();
9630         if (type < 0) {
9631                 pr_warn("failed to determine %s perf type: %s\n",
9632                         uprobe ? "uprobe" : "kprobe",
9633                         libbpf_strerror_r(type, errmsg, sizeof(errmsg)));
9634                 return type;
9635         }
9636         if (retprobe) {
9637                 int bit = uprobe ? determine_uprobe_retprobe_bit()
9638                                  : determine_kprobe_retprobe_bit();
9639
9640                 if (bit < 0) {
9641                         pr_warn("failed to determine %s retprobe bit: %s\n",
9642                                 uprobe ? "uprobe" : "kprobe",
9643                                 libbpf_strerror_r(bit, errmsg, sizeof(errmsg)));
9644                         return bit;
9645                 }
9646                 attr.config |= 1 << bit;
9647         }
9648         attr.size = sizeof(attr);
9649         attr.type = type;
9650         attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
9651         attr.config2 = offset;           /* kprobe_addr or probe_offset */
9652
9653         /* pid filter is meaningful only for uprobes */
9654         pfd = syscall(__NR_perf_event_open, &attr,
9655                       pid < 0 ? -1 : pid /* pid */,
9656                       pid == -1 ? 0 : -1 /* cpu */,
9657                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
9658         if (pfd < 0) {
9659                 err = -errno;
9660                 pr_warn("%s perf_event_open() failed: %s\n",
9661                         uprobe ? "uprobe" : "kprobe",
9662                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9663                 return err;
9664         }
9665         return pfd;
9666 }
9667
9668 struct bpf_link *bpf_program__attach_kprobe(struct bpf_program *prog,
9669                                             bool retprobe,
9670                                             const char *func_name)
9671 {
9672         char errmsg[STRERR_BUFSIZE];
9673         struct bpf_link *link;
9674         int pfd, err;
9675
9676         pfd = perf_event_open_probe(false /* uprobe */, retprobe, func_name,
9677                                     0 /* offset */, -1 /* pid */);
9678         if (pfd < 0) {
9679                 pr_warn("prog '%s': failed to create %s '%s' perf event: %s\n",
9680                         prog->name, retprobe ? "kretprobe" : "kprobe", func_name,
9681                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
9682                 return ERR_PTR(pfd);
9683         }
9684         link = bpf_program__attach_perf_event(prog, pfd);
9685         if (IS_ERR(link)) {
9686                 close(pfd);
9687                 err = PTR_ERR(link);
9688                 pr_warn("prog '%s': failed to attach to %s '%s': %s\n",
9689                         prog->name, retprobe ? "kretprobe" : "kprobe", func_name,
9690                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9691                 return link;
9692         }
9693         return link;
9694 }
9695
9696 static struct bpf_link *attach_kprobe(const struct bpf_sec_def *sec,
9697                                       struct bpf_program *prog)
9698 {
9699         const char *func_name;
9700         bool retprobe;
9701
9702         func_name = prog->sec_name + sec->len;
9703         retprobe = strcmp(sec->sec, "kretprobe/") == 0;
9704
9705         return bpf_program__attach_kprobe(prog, retprobe, func_name);
9706 }
9707
9708 struct bpf_link *bpf_program__attach_uprobe(struct bpf_program *prog,
9709                                             bool retprobe, pid_t pid,
9710                                             const char *binary_path,
9711                                             size_t func_offset)
9712 {
9713         char errmsg[STRERR_BUFSIZE];
9714         struct bpf_link *link;
9715         int pfd, err;
9716
9717         pfd = perf_event_open_probe(true /* uprobe */, retprobe,
9718                                     binary_path, func_offset, pid);
9719         if (pfd < 0) {
9720                 pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
9721                         prog->name, retprobe ? "uretprobe" : "uprobe",
9722                         binary_path, func_offset,
9723                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
9724                 return ERR_PTR(pfd);
9725         }
9726         link = bpf_program__attach_perf_event(prog, pfd);
9727         if (IS_ERR(link)) {
9728                 close(pfd);
9729                 err = PTR_ERR(link);
9730                 pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n",
9731                         prog->name, retprobe ? "uretprobe" : "uprobe",
9732                         binary_path, func_offset,
9733                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9734                 return link;
9735         }
9736         return link;
9737 }
9738
9739 static int determine_tracepoint_id(const char *tp_category,
9740                                    const char *tp_name)
9741 {
9742         char file[PATH_MAX];
9743         int ret;
9744
9745         ret = snprintf(file, sizeof(file),
9746                        "/sys/kernel/debug/tracing/events/%s/%s/id",
9747                        tp_category, tp_name);
9748         if (ret < 0)
9749                 return -errno;
9750         if (ret >= sizeof(file)) {
9751                 pr_debug("tracepoint %s/%s path is too long\n",
9752                          tp_category, tp_name);
9753                 return -E2BIG;
9754         }
9755         return parse_uint_from_file(file, "%d\n");
9756 }
9757
9758 static int perf_event_open_tracepoint(const char *tp_category,
9759                                       const char *tp_name)
9760 {
9761         struct perf_event_attr attr = {};
9762         char errmsg[STRERR_BUFSIZE];
9763         int tp_id, pfd, err;
9764
9765         tp_id = determine_tracepoint_id(tp_category, tp_name);
9766         if (tp_id < 0) {
9767                 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
9768                         tp_category, tp_name,
9769                         libbpf_strerror_r(tp_id, errmsg, sizeof(errmsg)));
9770                 return tp_id;
9771         }
9772
9773         attr.type = PERF_TYPE_TRACEPOINT;
9774         attr.size = sizeof(attr);
9775         attr.config = tp_id;
9776
9777         pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
9778                       -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
9779         if (pfd < 0) {
9780                 err = -errno;
9781                 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
9782                         tp_category, tp_name,
9783                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9784                 return err;
9785         }
9786         return pfd;
9787 }
9788
9789 struct bpf_link *bpf_program__attach_tracepoint(struct bpf_program *prog,
9790                                                 const char *tp_category,
9791                                                 const char *tp_name)
9792 {
9793         char errmsg[STRERR_BUFSIZE];
9794         struct bpf_link *link;
9795         int pfd, err;
9796
9797         pfd = perf_event_open_tracepoint(tp_category, tp_name);
9798         if (pfd < 0) {
9799                 pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
9800                         prog->name, tp_category, tp_name,
9801                         libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
9802                 return ERR_PTR(pfd);
9803         }
9804         link = bpf_program__attach_perf_event(prog, pfd);
9805         if (IS_ERR(link)) {
9806                 close(pfd);
9807                 err = PTR_ERR(link);
9808                 pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n",
9809                         prog->name, tp_category, tp_name,
9810                         libbpf_strerror_r(err, errmsg, sizeof(errmsg)));
9811                 return link;
9812         }
9813         return link;
9814 }
9815
9816 static struct bpf_link *attach_tp(const struct bpf_sec_def *sec,
9817                                   struct bpf_program *prog)
9818 {
9819         char *sec_name, *tp_cat, *tp_name;
9820         struct bpf_link *link;
9821
9822         sec_name = strdup(prog->sec_name);
9823         if (!sec_name)
9824                 return ERR_PTR(-ENOMEM);
9825
9826         /* extract "tp/<category>/<name>" */
9827         tp_cat = sec_name + sec->len;
9828         tp_name = strchr(tp_cat, '/');
9829         if (!tp_name) {
9830                 link = ERR_PTR(-EINVAL);
9831                 goto out;
9832         }
9833         *tp_name = '\0';
9834         tp_name++;
9835
9836         link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
9837 out:
9838         free(sec_name);
9839         return link;
9840 }
9841
9842 struct bpf_link *bpf_program__attach_raw_tracepoint(struct bpf_program *prog,
9843                                                     const char *tp_name)
9844 {
9845         char errmsg[STRERR_BUFSIZE];
9846         struct bpf_link *link;
9847         int prog_fd, pfd;
9848
9849         prog_fd = bpf_program__fd(prog);
9850         if (prog_fd < 0) {
9851                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
9852                 return ERR_PTR(-EINVAL);
9853         }
9854
9855         link = calloc(1, sizeof(*link));
9856         if (!link)
9857                 return ERR_PTR(-ENOMEM);
9858         link->detach = &bpf_link__detach_fd;
9859
9860         pfd = bpf_raw_tracepoint_open(tp_name, prog_fd);
9861         if (pfd < 0) {
9862                 pfd = -errno;
9863                 free(link);
9864                 pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n",
9865                         prog->name, tp_name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
9866                 return ERR_PTR(pfd);
9867         }
9868         link->fd = pfd;
9869         return link;
9870 }
9871
9872 static struct bpf_link *attach_raw_tp(const struct bpf_sec_def *sec,
9873                                       struct bpf_program *prog)
9874 {
9875         const char *tp_name = prog->sec_name + sec->len;
9876
9877         return bpf_program__attach_raw_tracepoint(prog, tp_name);
9878 }
9879
9880 /* Common logic for all BPF program types that attach to a btf_id */
9881 static struct bpf_link *bpf_program__attach_btf_id(struct bpf_program *prog)
9882 {
9883         char errmsg[STRERR_BUFSIZE];
9884         struct bpf_link *link;
9885         int prog_fd, pfd;
9886
9887         prog_fd = bpf_program__fd(prog);
9888         if (prog_fd < 0) {
9889                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
9890                 return ERR_PTR(-EINVAL);
9891         }
9892
9893         link = calloc(1, sizeof(*link));
9894         if (!link)
9895                 return ERR_PTR(-ENOMEM);
9896         link->detach = &bpf_link__detach_fd;
9897
9898         pfd = bpf_raw_tracepoint_open(NULL, prog_fd);
9899         if (pfd < 0) {
9900                 pfd = -errno;
9901                 free(link);
9902                 pr_warn("prog '%s': failed to attach: %s\n",
9903                         prog->name, libbpf_strerror_r(pfd, errmsg, sizeof(errmsg)));
9904                 return ERR_PTR(pfd);
9905         }
9906         link->fd = pfd;
9907         return (struct bpf_link *)link;
9908 }
9909
9910 struct bpf_link *bpf_program__attach_trace(struct bpf_program *prog)
9911 {
9912         return bpf_program__attach_btf_id(prog);
9913 }
9914
9915 struct bpf_link *bpf_program__attach_lsm(struct bpf_program *prog)
9916 {
9917         return bpf_program__attach_btf_id(prog);
9918 }
9919
9920 static struct bpf_link *attach_trace(const struct bpf_sec_def *sec,
9921                                      struct bpf_program *prog)
9922 {
9923         return bpf_program__attach_trace(prog);
9924 }
9925
9926 static struct bpf_link *attach_lsm(const struct bpf_sec_def *sec,
9927                                    struct bpf_program *prog)
9928 {
9929         return bpf_program__attach_lsm(prog);
9930 }
9931
9932 static struct bpf_link *attach_iter(const struct bpf_sec_def *sec,
9933                                     struct bpf_program *prog)
9934 {
9935         return bpf_program__attach_iter(prog, NULL);
9936 }
9937
9938 static struct bpf_link *
9939 bpf_program__attach_fd(struct bpf_program *prog, int target_fd, int btf_id,
9940                        const char *target_name)
9941 {
9942         DECLARE_LIBBPF_OPTS(bpf_link_create_opts, opts,
9943                             .target_btf_id = btf_id);
9944         enum bpf_attach_type attach_type;
9945         char errmsg[STRERR_BUFSIZE];
9946         struct bpf_link *link;
9947         int prog_fd, link_fd;
9948
9949         prog_fd = bpf_program__fd(prog);
9950         if (prog_fd < 0) {
9951                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
9952                 return ERR_PTR(-EINVAL);
9953         }
9954
9955         link = calloc(1, sizeof(*link));
9956         if (!link)
9957                 return ERR_PTR(-ENOMEM);
9958         link->detach = &bpf_link__detach_fd;
9959
9960         attach_type = bpf_program__get_expected_attach_type(prog);
9961         link_fd = bpf_link_create(prog_fd, target_fd, attach_type, &opts);
9962         if (link_fd < 0) {
9963                 link_fd = -errno;
9964                 free(link);
9965                 pr_warn("prog '%s': failed to attach to %s: %s\n",
9966                         prog->name, target_name,
9967                         libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
9968                 return ERR_PTR(link_fd);
9969         }
9970         link->fd = link_fd;
9971         return link;
9972 }
9973
9974 struct bpf_link *
9975 bpf_program__attach_cgroup(struct bpf_program *prog, int cgroup_fd)
9976 {
9977         return bpf_program__attach_fd(prog, cgroup_fd, 0, "cgroup");
9978 }
9979
9980 struct bpf_link *
9981 bpf_program__attach_netns(struct bpf_program *prog, int netns_fd)
9982 {
9983         return bpf_program__attach_fd(prog, netns_fd, 0, "netns");
9984 }
9985
9986 struct bpf_link *bpf_program__attach_xdp(struct bpf_program *prog, int ifindex)
9987 {
9988         /* target_fd/target_ifindex use the same field in LINK_CREATE */
9989         return bpf_program__attach_fd(prog, ifindex, 0, "xdp");
9990 }
9991
9992 struct bpf_link *bpf_program__attach_freplace(struct bpf_program *prog,
9993                                               int target_fd,
9994                                               const char *attach_func_name)
9995 {
9996         int btf_id;
9997
9998         if (!!target_fd != !!attach_func_name) {
9999                 pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n",
10000                         prog->name);
10001                 return ERR_PTR(-EINVAL);
10002         }
10003
10004         if (prog->type != BPF_PROG_TYPE_EXT) {
10005                 pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace",
10006                         prog->name);
10007                 return ERR_PTR(-EINVAL);
10008         }
10009
10010         if (target_fd) {
10011                 btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd);
10012                 if (btf_id < 0)
10013                         return ERR_PTR(btf_id);
10014
10015                 return bpf_program__attach_fd(prog, target_fd, btf_id, "freplace");
10016         } else {
10017                 /* no target, so use raw_tracepoint_open for compatibility
10018                  * with old kernels
10019                  */
10020                 return bpf_program__attach_trace(prog);
10021         }
10022 }
10023
10024 struct bpf_link *
10025 bpf_program__attach_iter(struct bpf_program *prog,
10026                          const struct bpf_iter_attach_opts *opts)
10027 {
10028         DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
10029         char errmsg[STRERR_BUFSIZE];
10030         struct bpf_link *link;
10031         int prog_fd, link_fd;
10032         __u32 target_fd = 0;
10033
10034         if (!OPTS_VALID(opts, bpf_iter_attach_opts))
10035                 return ERR_PTR(-EINVAL);
10036
10037         link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0);
10038         link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0);
10039
10040         prog_fd = bpf_program__fd(prog);
10041         if (prog_fd < 0) {
10042                 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
10043                 return ERR_PTR(-EINVAL);
10044         }
10045
10046         link = calloc(1, sizeof(*link));
10047         if (!link)
10048                 return ERR_PTR(-ENOMEM);
10049         link->detach = &bpf_link__detach_fd;
10050
10051         link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER,
10052                                   &link_create_opts);
10053         if (link_fd < 0) {
10054                 link_fd = -errno;
10055                 free(link);
10056                 pr_warn("prog '%s': failed to attach to iterator: %s\n",
10057                         prog->name, libbpf_strerror_r(link_fd, errmsg, sizeof(errmsg)));
10058                 return ERR_PTR(link_fd);
10059         }
10060         link->fd = link_fd;
10061         return link;
10062 }
10063
10064 struct bpf_link *bpf_program__attach(struct bpf_program *prog)
10065 {
10066         const struct bpf_sec_def *sec_def;
10067
10068         sec_def = find_sec_def(prog->sec_name);
10069         if (!sec_def || !sec_def->attach_fn)
10070                 return ERR_PTR(-ESRCH);
10071
10072         return sec_def->attach_fn(sec_def, prog);
10073 }
10074
10075 static int bpf_link__detach_struct_ops(struct bpf_link *link)
10076 {
10077         __u32 zero = 0;
10078
10079         if (bpf_map_delete_elem(link->fd, &zero))
10080                 return -errno;
10081
10082         return 0;
10083 }
10084
10085 struct bpf_link *bpf_map__attach_struct_ops(struct bpf_map *map)
10086 {
10087         struct bpf_struct_ops *st_ops;
10088         struct bpf_link *link;
10089         __u32 i, zero = 0;
10090         int err;
10091
10092         if (!bpf_map__is_struct_ops(map) || map->fd == -1)
10093                 return ERR_PTR(-EINVAL);
10094
10095         link = calloc(1, sizeof(*link));
10096         if (!link)
10097                 return ERR_PTR(-EINVAL);
10098
10099         st_ops = map->st_ops;
10100         for (i = 0; i < btf_vlen(st_ops->type); i++) {
10101                 struct bpf_program *prog = st_ops->progs[i];
10102                 void *kern_data;
10103                 int prog_fd;
10104
10105                 if (!prog)
10106                         continue;
10107
10108                 prog_fd = bpf_program__fd(prog);
10109                 kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
10110                 *(unsigned long *)kern_data = prog_fd;
10111         }
10112
10113         err = bpf_map_update_elem(map->fd, &zero, st_ops->kern_vdata, 0);
10114         if (err) {
10115                 err = -errno;
10116                 free(link);
10117                 return ERR_PTR(err);
10118         }
10119
10120         link->detach = bpf_link__detach_struct_ops;
10121         link->fd = map->fd;
10122
10123         return link;
10124 }
10125
10126 enum bpf_perf_event_ret
10127 bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
10128                            void **copy_mem, size_t *copy_size,
10129                            bpf_perf_event_print_t fn, void *private_data)
10130 {
10131         struct perf_event_mmap_page *header = mmap_mem;
10132         __u64 data_head = ring_buffer_read_head(header);
10133         __u64 data_tail = header->data_tail;
10134         void *base = ((__u8 *)header) + page_size;
10135         int ret = LIBBPF_PERF_EVENT_CONT;
10136         struct perf_event_header *ehdr;
10137         size_t ehdr_size;
10138
10139         while (data_head != data_tail) {
10140                 ehdr = base + (data_tail & (mmap_size - 1));
10141                 ehdr_size = ehdr->size;
10142
10143                 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
10144                         void *copy_start = ehdr;
10145                         size_t len_first = base + mmap_size - copy_start;
10146                         size_t len_secnd = ehdr_size - len_first;
10147
10148                         if (*copy_size < ehdr_size) {
10149                                 free(*copy_mem);
10150                                 *copy_mem = malloc(ehdr_size);
10151                                 if (!*copy_mem) {
10152                                         *copy_size = 0;
10153                                         ret = LIBBPF_PERF_EVENT_ERROR;
10154                                         break;
10155                                 }
10156                                 *copy_size = ehdr_size;
10157                         }
10158
10159                         memcpy(*copy_mem, copy_start, len_first);
10160                         memcpy(*copy_mem + len_first, base, len_secnd);
10161                         ehdr = *copy_mem;
10162                 }
10163
10164                 ret = fn(ehdr, private_data);
10165                 data_tail += ehdr_size;
10166                 if (ret != LIBBPF_PERF_EVENT_CONT)
10167                         break;
10168         }
10169
10170         ring_buffer_write_tail(header, data_tail);
10171         return ret;
10172 }
10173
10174 struct perf_buffer;
10175
10176 struct perf_buffer_params {
10177         struct perf_event_attr *attr;
10178         /* if event_cb is specified, it takes precendence */
10179         perf_buffer_event_fn event_cb;
10180         /* sample_cb and lost_cb are higher-level common-case callbacks */
10181         perf_buffer_sample_fn sample_cb;
10182         perf_buffer_lost_fn lost_cb;
10183         void *ctx;
10184         int cpu_cnt;
10185         int *cpus;
10186         int *map_keys;
10187 };
10188
10189 struct perf_cpu_buf {
10190         struct perf_buffer *pb;
10191         void *base; /* mmap()'ed memory */
10192         void *buf; /* for reconstructing segmented data */
10193         size_t buf_size;
10194         int fd;
10195         int cpu;
10196         int map_key;
10197 };
10198
10199 struct perf_buffer {
10200         perf_buffer_event_fn event_cb;
10201         perf_buffer_sample_fn sample_cb;
10202         perf_buffer_lost_fn lost_cb;
10203         void *ctx; /* passed into callbacks */
10204
10205         size_t page_size;
10206         size_t mmap_size;
10207         struct perf_cpu_buf **cpu_bufs;
10208         struct epoll_event *events;
10209         int cpu_cnt; /* number of allocated CPU buffers */
10210         int epoll_fd; /* perf event FD */
10211         int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
10212 };
10213
10214 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
10215                                       struct perf_cpu_buf *cpu_buf)
10216 {
10217         if (!cpu_buf)
10218                 return;
10219         if (cpu_buf->base &&
10220             munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
10221                 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
10222         if (cpu_buf->fd >= 0) {
10223                 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
10224                 close(cpu_buf->fd);
10225         }
10226         free(cpu_buf->buf);
10227         free(cpu_buf);
10228 }
10229
10230 void perf_buffer__free(struct perf_buffer *pb)
10231 {
10232         int i;
10233
10234         if (IS_ERR_OR_NULL(pb))
10235                 return;
10236         if (pb->cpu_bufs) {
10237                 for (i = 0; i < pb->cpu_cnt; i++) {
10238                         struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
10239
10240                         if (!cpu_buf)
10241                                 continue;
10242
10243                         bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
10244                         perf_buffer__free_cpu_buf(pb, cpu_buf);
10245                 }
10246                 free(pb->cpu_bufs);
10247         }
10248         if (pb->epoll_fd >= 0)
10249                 close(pb->epoll_fd);
10250         free(pb->events);
10251         free(pb);
10252 }
10253
10254 static struct perf_cpu_buf *
10255 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
10256                           int cpu, int map_key)
10257 {
10258         struct perf_cpu_buf *cpu_buf;
10259         char msg[STRERR_BUFSIZE];
10260         int err;
10261
10262         cpu_buf = calloc(1, sizeof(*cpu_buf));
10263         if (!cpu_buf)
10264                 return ERR_PTR(-ENOMEM);
10265
10266         cpu_buf->pb = pb;
10267         cpu_buf->cpu = cpu;
10268         cpu_buf->map_key = map_key;
10269
10270         cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
10271                               -1, PERF_FLAG_FD_CLOEXEC);
10272         if (cpu_buf->fd < 0) {
10273                 err = -errno;
10274                 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
10275                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
10276                 goto error;
10277         }
10278
10279         cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
10280                              PROT_READ | PROT_WRITE, MAP_SHARED,
10281                              cpu_buf->fd, 0);
10282         if (cpu_buf->base == MAP_FAILED) {
10283                 cpu_buf->base = NULL;
10284                 err = -errno;
10285                 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
10286                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
10287                 goto error;
10288         }
10289
10290         if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
10291                 err = -errno;
10292                 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
10293                         cpu, libbpf_strerror_r(err, msg, sizeof(msg)));
10294                 goto error;
10295         }
10296
10297         return cpu_buf;
10298
10299 error:
10300         perf_buffer__free_cpu_buf(pb, cpu_buf);
10301         return (struct perf_cpu_buf *)ERR_PTR(err);
10302 }
10303
10304 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
10305                                               struct perf_buffer_params *p);
10306
10307 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
10308                                      const struct perf_buffer_opts *opts)
10309 {
10310         struct perf_buffer_params p = {};
10311         struct perf_event_attr attr = { 0, };
10312
10313         attr.config = PERF_COUNT_SW_BPF_OUTPUT;
10314         attr.type = PERF_TYPE_SOFTWARE;
10315         attr.sample_type = PERF_SAMPLE_RAW;
10316         attr.sample_period = 1;
10317         attr.wakeup_events = 1;
10318
10319         p.attr = &attr;
10320         p.sample_cb = opts ? opts->sample_cb : NULL;
10321         p.lost_cb = opts ? opts->lost_cb : NULL;
10322         p.ctx = opts ? opts->ctx : NULL;
10323
10324         return __perf_buffer__new(map_fd, page_cnt, &p);
10325 }
10326
10327 struct perf_buffer *
10328 perf_buffer__new_raw(int map_fd, size_t page_cnt,
10329                      const struct perf_buffer_raw_opts *opts)
10330 {
10331         struct perf_buffer_params p = {};
10332
10333         p.attr = opts->attr;
10334         p.event_cb = opts->event_cb;
10335         p.ctx = opts->ctx;
10336         p.cpu_cnt = opts->cpu_cnt;
10337         p.cpus = opts->cpus;
10338         p.map_keys = opts->map_keys;
10339
10340         return __perf_buffer__new(map_fd, page_cnt, &p);
10341 }
10342
10343 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
10344                                               struct perf_buffer_params *p)
10345 {
10346         const char *online_cpus_file = "/sys/devices/system/cpu/online";
10347         struct bpf_map_info map;
10348         char msg[STRERR_BUFSIZE];
10349         struct perf_buffer *pb;
10350         bool *online = NULL;
10351         __u32 map_info_len;
10352         int err, i, j, n;
10353
10354         if (page_cnt & (page_cnt - 1)) {
10355                 pr_warn("page count should be power of two, but is %zu\n",
10356                         page_cnt);
10357                 return ERR_PTR(-EINVAL);
10358         }
10359
10360         /* best-effort sanity checks */
10361         memset(&map, 0, sizeof(map));
10362         map_info_len = sizeof(map);
10363         err = bpf_obj_get_info_by_fd(map_fd, &map, &map_info_len);
10364         if (err) {
10365                 err = -errno;
10366                 /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return
10367                  * -EBADFD, -EFAULT, or -E2BIG on real error
10368                  */
10369                 if (err != -EINVAL) {
10370                         pr_warn("failed to get map info for map FD %d: %s\n",
10371                                 map_fd, libbpf_strerror_r(err, msg, sizeof(msg)));
10372                         return ERR_PTR(err);
10373                 }
10374                 pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n",
10375                          map_fd);
10376         } else {
10377                 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
10378                         pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
10379                                 map.name);
10380                         return ERR_PTR(-EINVAL);
10381                 }
10382         }
10383
10384         pb = calloc(1, sizeof(*pb));
10385         if (!pb)
10386                 return ERR_PTR(-ENOMEM);
10387
10388         pb->event_cb = p->event_cb;
10389         pb->sample_cb = p->sample_cb;
10390         pb->lost_cb = p->lost_cb;
10391         pb->ctx = p->ctx;
10392
10393         pb->page_size = getpagesize();
10394         pb->mmap_size = pb->page_size * page_cnt;
10395         pb->map_fd = map_fd;
10396
10397         pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
10398         if (pb->epoll_fd < 0) {
10399                 err = -errno;
10400                 pr_warn("failed to create epoll instance: %s\n",
10401                         libbpf_strerror_r(err, msg, sizeof(msg)));
10402                 goto error;
10403         }
10404
10405         if (p->cpu_cnt > 0) {
10406                 pb->cpu_cnt = p->cpu_cnt;
10407         } else {
10408                 pb->cpu_cnt = libbpf_num_possible_cpus();
10409                 if (pb->cpu_cnt < 0) {
10410                         err = pb->cpu_cnt;
10411                         goto error;
10412                 }
10413                 if (map.max_entries && map.max_entries < pb->cpu_cnt)
10414                         pb->cpu_cnt = map.max_entries;
10415         }
10416
10417         pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
10418         if (!pb->events) {
10419                 err = -ENOMEM;
10420                 pr_warn("failed to allocate events: out of memory\n");
10421                 goto error;
10422         }
10423         pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
10424         if (!pb->cpu_bufs) {
10425                 err = -ENOMEM;
10426                 pr_warn("failed to allocate buffers: out of memory\n");
10427                 goto error;
10428         }
10429
10430         err = parse_cpu_mask_file(online_cpus_file, &online, &n);
10431         if (err) {
10432                 pr_warn("failed to get online CPU mask: %d\n", err);
10433                 goto error;
10434         }
10435
10436         for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
10437                 struct perf_cpu_buf *cpu_buf;
10438                 int cpu, map_key;
10439
10440                 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
10441                 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
10442
10443                 /* in case user didn't explicitly requested particular CPUs to
10444                  * be attached to, skip offline/not present CPUs
10445                  */
10446                 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
10447                         continue;
10448
10449                 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
10450                 if (IS_ERR(cpu_buf)) {
10451                         err = PTR_ERR(cpu_buf);
10452                         goto error;
10453                 }
10454
10455                 pb->cpu_bufs[j] = cpu_buf;
10456
10457                 err = bpf_map_update_elem(pb->map_fd, &map_key,
10458                                           &cpu_buf->fd, 0);
10459                 if (err) {
10460                         err = -errno;
10461                         pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
10462                                 cpu, map_key, cpu_buf->fd,
10463                                 libbpf_strerror_r(err, msg, sizeof(msg)));
10464                         goto error;
10465                 }
10466
10467                 pb->events[j].events = EPOLLIN;
10468                 pb->events[j].data.ptr = cpu_buf;
10469                 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
10470                               &pb->events[j]) < 0) {
10471                         err = -errno;
10472                         pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
10473                                 cpu, cpu_buf->fd,
10474                                 libbpf_strerror_r(err, msg, sizeof(msg)));
10475                         goto error;
10476                 }
10477                 j++;
10478         }
10479         pb->cpu_cnt = j;
10480         free(online);
10481
10482         return pb;
10483
10484 error:
10485         free(online);
10486         if (pb)
10487                 perf_buffer__free(pb);
10488         return ERR_PTR(err);
10489 }
10490
10491 struct perf_sample_raw {
10492         struct perf_event_header header;
10493         uint32_t size;
10494         char data[];
10495 };
10496
10497 struct perf_sample_lost {
10498         struct perf_event_header header;
10499         uint64_t id;
10500         uint64_t lost;
10501         uint64_t sample_id;
10502 };
10503
10504 static enum bpf_perf_event_ret
10505 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
10506 {
10507         struct perf_cpu_buf *cpu_buf = ctx;
10508         struct perf_buffer *pb = cpu_buf->pb;
10509         void *data = e;
10510
10511         /* user wants full control over parsing perf event */
10512         if (pb->event_cb)
10513                 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
10514
10515         switch (e->type) {
10516         case PERF_RECORD_SAMPLE: {
10517                 struct perf_sample_raw *s = data;
10518
10519                 if (pb->sample_cb)
10520                         pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
10521                 break;
10522         }
10523         case PERF_RECORD_LOST: {
10524                 struct perf_sample_lost *s = data;
10525
10526                 if (pb->lost_cb)
10527                         pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
10528                 break;
10529         }
10530         default:
10531                 pr_warn("unknown perf sample type %d\n", e->type);
10532                 return LIBBPF_PERF_EVENT_ERROR;
10533         }
10534         return LIBBPF_PERF_EVENT_CONT;
10535 }
10536
10537 static int perf_buffer__process_records(struct perf_buffer *pb,
10538                                         struct perf_cpu_buf *cpu_buf)
10539 {
10540         enum bpf_perf_event_ret ret;
10541
10542         ret = bpf_perf_event_read_simple(cpu_buf->base, pb->mmap_size,
10543                                          pb->page_size, &cpu_buf->buf,
10544                                          &cpu_buf->buf_size,
10545                                          perf_buffer__process_record, cpu_buf);
10546         if (ret != LIBBPF_PERF_EVENT_CONT)
10547                 return ret;
10548         return 0;
10549 }
10550
10551 int perf_buffer__epoll_fd(const struct perf_buffer *pb)
10552 {
10553         return pb->epoll_fd;
10554 }
10555
10556 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
10557 {
10558         int i, cnt, err;
10559
10560         cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
10561         for (i = 0; i < cnt; i++) {
10562                 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
10563
10564                 err = perf_buffer__process_records(pb, cpu_buf);
10565                 if (err) {
10566                         pr_warn("error while processing records: %d\n", err);
10567                         return err;
10568                 }
10569         }
10570         return cnt < 0 ? -errno : cnt;
10571 }
10572
10573 /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer
10574  * manager.
10575  */
10576 size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb)
10577 {
10578         return pb->cpu_cnt;
10579 }
10580
10581 /*
10582  * Return perf_event FD of a ring buffer in *buf_idx* slot of
10583  * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using
10584  * select()/poll()/epoll() Linux syscalls.
10585  */
10586 int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx)
10587 {
10588         struct perf_cpu_buf *cpu_buf;
10589
10590         if (buf_idx >= pb->cpu_cnt)
10591                 return -EINVAL;
10592
10593         cpu_buf = pb->cpu_bufs[buf_idx];
10594         if (!cpu_buf)
10595                 return -ENOENT;
10596
10597         return cpu_buf->fd;
10598 }
10599
10600 /*
10601  * Consume data from perf ring buffer corresponding to slot *buf_idx* in
10602  * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to
10603  * consume, do nothing and return success.
10604  * Returns:
10605  *   - 0 on success;
10606  *   - <0 on failure.
10607  */
10608 int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx)
10609 {
10610         struct perf_cpu_buf *cpu_buf;
10611
10612         if (buf_idx >= pb->cpu_cnt)
10613                 return -EINVAL;
10614
10615         cpu_buf = pb->cpu_bufs[buf_idx];
10616         if (!cpu_buf)
10617                 return -ENOENT;
10618
10619         return perf_buffer__process_records(pb, cpu_buf);
10620 }
10621
10622 int perf_buffer__consume(struct perf_buffer *pb)
10623 {
10624         int i, err;
10625
10626         for (i = 0; i < pb->cpu_cnt; i++) {
10627                 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
10628
10629                 if (!cpu_buf)
10630                         continue;
10631
10632                 err = perf_buffer__process_records(pb, cpu_buf);
10633                 if (err) {
10634                         pr_warn("perf_buffer: failed to process records in buffer #%d: %d\n", i, err);
10635                         return err;
10636                 }
10637         }
10638         return 0;
10639 }
10640
10641 struct bpf_prog_info_array_desc {
10642         int     array_offset;   /* e.g. offset of jited_prog_insns */
10643         int     count_offset;   /* e.g. offset of jited_prog_len */
10644         int     size_offset;    /* > 0: offset of rec size,
10645                                  * < 0: fix size of -size_offset
10646                                  */
10647 };
10648
10649 static struct bpf_prog_info_array_desc bpf_prog_info_array_desc[] = {
10650         [BPF_PROG_INFO_JITED_INSNS] = {
10651                 offsetof(struct bpf_prog_info, jited_prog_insns),
10652                 offsetof(struct bpf_prog_info, jited_prog_len),
10653                 -1,
10654         },
10655         [BPF_PROG_INFO_XLATED_INSNS] = {
10656                 offsetof(struct bpf_prog_info, xlated_prog_insns),
10657                 offsetof(struct bpf_prog_info, xlated_prog_len),
10658                 -1,
10659         },
10660         [BPF_PROG_INFO_MAP_IDS] = {
10661                 offsetof(struct bpf_prog_info, map_ids),
10662                 offsetof(struct bpf_prog_info, nr_map_ids),
10663                 -(int)sizeof(__u32),
10664         },
10665         [BPF_PROG_INFO_JITED_KSYMS] = {
10666                 offsetof(struct bpf_prog_info, jited_ksyms),
10667                 offsetof(struct bpf_prog_info, nr_jited_ksyms),
10668                 -(int)sizeof(__u64),
10669         },
10670         [BPF_PROG_INFO_JITED_FUNC_LENS] = {
10671                 offsetof(struct bpf_prog_info, jited_func_lens),
10672                 offsetof(struct bpf_prog_info, nr_jited_func_lens),
10673                 -(int)sizeof(__u32),
10674         },
10675         [BPF_PROG_INFO_FUNC_INFO] = {
10676                 offsetof(struct bpf_prog_info, func_info),
10677                 offsetof(struct bpf_prog_info, nr_func_info),
10678                 offsetof(struct bpf_prog_info, func_info_rec_size),
10679         },
10680         [BPF_PROG_INFO_LINE_INFO] = {
10681                 offsetof(struct bpf_prog_info, line_info),
10682                 offsetof(struct bpf_prog_info, nr_line_info),
10683                 offsetof(struct bpf_prog_info, line_info_rec_size),
10684         },
10685         [BPF_PROG_INFO_JITED_LINE_INFO] = {
10686                 offsetof(struct bpf_prog_info, jited_line_info),
10687                 offsetof(struct bpf_prog_info, nr_jited_line_info),
10688                 offsetof(struct bpf_prog_info, jited_line_info_rec_size),
10689         },
10690         [BPF_PROG_INFO_PROG_TAGS] = {
10691                 offsetof(struct bpf_prog_info, prog_tags),
10692                 offsetof(struct bpf_prog_info, nr_prog_tags),
10693                 -(int)sizeof(__u8) * BPF_TAG_SIZE,
10694         },
10695
10696 };
10697
10698 static __u32 bpf_prog_info_read_offset_u32(struct bpf_prog_info *info,
10699                                            int offset)
10700 {
10701         __u32 *array = (__u32 *)info;
10702
10703         if (offset >= 0)
10704                 return array[offset / sizeof(__u32)];
10705         return -(int)offset;
10706 }
10707
10708 static __u64 bpf_prog_info_read_offset_u64(struct bpf_prog_info *info,
10709                                            int offset)
10710 {
10711         __u64 *array = (__u64 *)info;
10712
10713         if (offset >= 0)
10714                 return array[offset / sizeof(__u64)];
10715         return -(int)offset;
10716 }
10717
10718 static void bpf_prog_info_set_offset_u32(struct bpf_prog_info *info, int offset,
10719                                          __u32 val)
10720 {
10721         __u32 *array = (__u32 *)info;
10722
10723         if (offset >= 0)
10724                 array[offset / sizeof(__u32)] = val;
10725 }
10726
10727 static void bpf_prog_info_set_offset_u64(struct bpf_prog_info *info, int offset,
10728                                          __u64 val)
10729 {
10730         __u64 *array = (__u64 *)info;
10731
10732         if (offset >= 0)
10733                 array[offset / sizeof(__u64)] = val;
10734 }
10735
10736 struct bpf_prog_info_linear *
10737 bpf_program__get_prog_info_linear(int fd, __u64 arrays)
10738 {
10739         struct bpf_prog_info_linear *info_linear;
10740         struct bpf_prog_info info = {};
10741         __u32 info_len = sizeof(info);
10742         __u32 data_len = 0;
10743         int i, err;
10744         void *ptr;
10745
10746         if (arrays >> BPF_PROG_INFO_LAST_ARRAY)
10747                 return ERR_PTR(-EINVAL);
10748
10749         /* step 1: get array dimensions */
10750         err = bpf_obj_get_info_by_fd(fd, &info, &info_len);
10751         if (err) {
10752                 pr_debug("can't get prog info: %s", strerror(errno));
10753                 return ERR_PTR(-EFAULT);
10754         }
10755
10756         /* step 2: calculate total size of all arrays */
10757         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
10758                 bool include_array = (arrays & (1UL << i)) > 0;
10759                 struct bpf_prog_info_array_desc *desc;
10760                 __u32 count, size;
10761
10762                 desc = bpf_prog_info_array_desc + i;
10763
10764                 /* kernel is too old to support this field */
10765                 if (info_len < desc->array_offset + sizeof(__u32) ||
10766                     info_len < desc->count_offset + sizeof(__u32) ||
10767                     (desc->size_offset > 0 && info_len < desc->size_offset))
10768                         include_array = false;
10769
10770                 if (!include_array) {
10771                         arrays &= ~(1UL << i);  /* clear the bit */
10772                         continue;
10773                 }
10774
10775                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
10776                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
10777
10778                 data_len += count * size;
10779         }
10780
10781         /* step 3: allocate continuous memory */
10782         data_len = roundup(data_len, sizeof(__u64));
10783         info_linear = malloc(sizeof(struct bpf_prog_info_linear) + data_len);
10784         if (!info_linear)
10785                 return ERR_PTR(-ENOMEM);
10786
10787         /* step 4: fill data to info_linear->info */
10788         info_linear->arrays = arrays;
10789         memset(&info_linear->info, 0, sizeof(info));
10790         ptr = info_linear->data;
10791
10792         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
10793                 struct bpf_prog_info_array_desc *desc;
10794                 __u32 count, size;
10795
10796                 if ((arrays & (1UL << i)) == 0)
10797                         continue;
10798
10799                 desc  = bpf_prog_info_array_desc + i;
10800                 count = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
10801                 size  = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
10802                 bpf_prog_info_set_offset_u32(&info_linear->info,
10803                                              desc->count_offset, count);
10804                 bpf_prog_info_set_offset_u32(&info_linear->info,
10805                                              desc->size_offset, size);
10806                 bpf_prog_info_set_offset_u64(&info_linear->info,
10807                                              desc->array_offset,
10808                                              ptr_to_u64(ptr));
10809                 ptr += count * size;
10810         }
10811
10812         /* step 5: call syscall again to get required arrays */
10813         err = bpf_obj_get_info_by_fd(fd, &info_linear->info, &info_len);
10814         if (err) {
10815                 pr_debug("can't get prog info: %s", strerror(errno));
10816                 free(info_linear);
10817                 return ERR_PTR(-EFAULT);
10818         }
10819
10820         /* step 6: verify the data */
10821         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
10822                 struct bpf_prog_info_array_desc *desc;
10823                 __u32 v1, v2;
10824
10825                 if ((arrays & (1UL << i)) == 0)
10826                         continue;
10827
10828                 desc = bpf_prog_info_array_desc + i;
10829                 v1 = bpf_prog_info_read_offset_u32(&info, desc->count_offset);
10830                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
10831                                                    desc->count_offset);
10832                 if (v1 != v2)
10833                         pr_warn("%s: mismatch in element count\n", __func__);
10834
10835                 v1 = bpf_prog_info_read_offset_u32(&info, desc->size_offset);
10836                 v2 = bpf_prog_info_read_offset_u32(&info_linear->info,
10837                                                    desc->size_offset);
10838                 if (v1 != v2)
10839                         pr_warn("%s: mismatch in rec size\n", __func__);
10840         }
10841
10842         /* step 7: update info_len and data_len */
10843         info_linear->info_len = sizeof(struct bpf_prog_info);
10844         info_linear->data_len = data_len;
10845
10846         return info_linear;
10847 }
10848
10849 void bpf_program__bpil_addr_to_offs(struct bpf_prog_info_linear *info_linear)
10850 {
10851         int i;
10852
10853         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
10854                 struct bpf_prog_info_array_desc *desc;
10855                 __u64 addr, offs;
10856
10857                 if ((info_linear->arrays & (1UL << i)) == 0)
10858                         continue;
10859
10860                 desc = bpf_prog_info_array_desc + i;
10861                 addr = bpf_prog_info_read_offset_u64(&info_linear->info,
10862                                                      desc->array_offset);
10863                 offs = addr - ptr_to_u64(info_linear->data);
10864                 bpf_prog_info_set_offset_u64(&info_linear->info,
10865                                              desc->array_offset, offs);
10866         }
10867 }
10868
10869 void bpf_program__bpil_offs_to_addr(struct bpf_prog_info_linear *info_linear)
10870 {
10871         int i;
10872
10873         for (i = BPF_PROG_INFO_FIRST_ARRAY; i < BPF_PROG_INFO_LAST_ARRAY; ++i) {
10874                 struct bpf_prog_info_array_desc *desc;
10875                 __u64 addr, offs;
10876
10877                 if ((info_linear->arrays & (1UL << i)) == 0)
10878                         continue;
10879
10880                 desc = bpf_prog_info_array_desc + i;
10881                 offs = bpf_prog_info_read_offset_u64(&info_linear->info,
10882                                                      desc->array_offset);
10883                 addr = offs + ptr_to_u64(info_linear->data);
10884                 bpf_prog_info_set_offset_u64(&info_linear->info,
10885                                              desc->array_offset, addr);
10886         }
10887 }
10888
10889 int bpf_program__set_attach_target(struct bpf_program *prog,
10890                                    int attach_prog_fd,
10891                                    const char *attach_func_name)
10892 {
10893         int btf_obj_fd = 0, btf_id = 0, err;
10894
10895         if (!prog || attach_prog_fd < 0 || !attach_func_name)
10896                 return -EINVAL;
10897
10898         if (prog->obj->loaded)
10899                 return -EINVAL;
10900
10901         if (attach_prog_fd) {
10902                 btf_id = libbpf_find_prog_btf_id(attach_func_name,
10903                                                  attach_prog_fd);
10904                 if (btf_id < 0)
10905                         return btf_id;
10906         } else {
10907                 /* load btf_vmlinux, if not yet */
10908                 err = bpf_object__load_vmlinux_btf(prog->obj, true);
10909                 if (err)
10910                         return err;
10911                 err = find_kernel_btf_id(prog->obj, attach_func_name,
10912                                          prog->expected_attach_type,
10913                                          &btf_obj_fd, &btf_id);
10914                 if (err)
10915                         return err;
10916         }
10917
10918         prog->attach_btf_id = btf_id;
10919         prog->attach_btf_obj_fd = btf_obj_fd;
10920         prog->attach_prog_fd = attach_prog_fd;
10921         return 0;
10922 }
10923
10924 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
10925 {
10926         int err = 0, n, len, start, end = -1;
10927         bool *tmp;
10928
10929         *mask = NULL;
10930         *mask_sz = 0;
10931
10932         /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
10933         while (*s) {
10934                 if (*s == ',' || *s == '\n') {
10935                         s++;
10936                         continue;
10937                 }
10938                 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
10939                 if (n <= 0 || n > 2) {
10940                         pr_warn("Failed to get CPU range %s: %d\n", s, n);
10941                         err = -EINVAL;
10942                         goto cleanup;
10943                 } else if (n == 1) {
10944                         end = start;
10945                 }
10946                 if (start < 0 || start > end) {
10947                         pr_warn("Invalid CPU range [%d,%d] in %s\n",
10948                                 start, end, s);
10949                         err = -EINVAL;
10950                         goto cleanup;
10951                 }
10952                 tmp = realloc(*mask, end + 1);
10953                 if (!tmp) {
10954                         err = -ENOMEM;
10955                         goto cleanup;
10956                 }
10957                 *mask = tmp;
10958                 memset(tmp + *mask_sz, 0, start - *mask_sz);
10959                 memset(tmp + start, 1, end - start + 1);
10960                 *mask_sz = end + 1;
10961                 s += len;
10962         }
10963         if (!*mask_sz) {
10964                 pr_warn("Empty CPU range\n");
10965                 return -EINVAL;
10966         }
10967         return 0;
10968 cleanup:
10969         free(*mask);
10970         *mask = NULL;
10971         return err;
10972 }
10973
10974 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
10975 {
10976         int fd, err = 0, len;
10977         char buf[128];
10978
10979         fd = open(fcpu, O_RDONLY);
10980         if (fd < 0) {
10981                 err = -errno;
10982                 pr_warn("Failed to open cpu mask file %s: %d\n", fcpu, err);
10983                 return err;
10984         }
10985         len = read(fd, buf, sizeof(buf));
10986         close(fd);
10987         if (len <= 0) {
10988                 err = len ? -errno : -EINVAL;
10989                 pr_warn("Failed to read cpu mask from %s: %d\n", fcpu, err);
10990                 return err;
10991         }
10992         if (len >= sizeof(buf)) {
10993                 pr_warn("CPU mask is too big in file %s\n", fcpu);
10994                 return -E2BIG;
10995         }
10996         buf[len] = '\0';
10997
10998         return parse_cpu_mask_str(buf, mask, mask_sz);
10999 }
11000
11001 int libbpf_num_possible_cpus(void)
11002 {
11003         static const char *fcpu = "/sys/devices/system/cpu/possible";
11004         static int cpus;
11005         int err, n, i, tmp_cpus;
11006         bool *mask;
11007
11008         tmp_cpus = READ_ONCE(cpus);
11009         if (tmp_cpus > 0)
11010                 return tmp_cpus;
11011
11012         err = parse_cpu_mask_file(fcpu, &mask, &n);
11013         if (err)
11014                 return err;
11015
11016         tmp_cpus = 0;
11017         for (i = 0; i < n; i++) {
11018                 if (mask[i])
11019                         tmp_cpus++;
11020         }
11021         free(mask);
11022
11023         WRITE_ONCE(cpus, tmp_cpus);
11024         return tmp_cpus;
11025 }
11026
11027 int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
11028                               const struct bpf_object_open_opts *opts)
11029 {
11030         DECLARE_LIBBPF_OPTS(bpf_object_open_opts, skel_opts,
11031                 .object_name = s->name,
11032         );
11033         struct bpf_object *obj;
11034         int i;
11035
11036         /* Attempt to preserve opts->object_name, unless overriden by user
11037          * explicitly. Overwriting object name for skeletons is discouraged,
11038          * as it breaks global data maps, because they contain object name
11039          * prefix as their own map name prefix. When skeleton is generated,
11040          * bpftool is making an assumption that this name will stay the same.
11041          */
11042         if (opts) {
11043                 memcpy(&skel_opts, opts, sizeof(*opts));
11044                 if (!opts->object_name)
11045                         skel_opts.object_name = s->name;
11046         }
11047
11048         obj = bpf_object__open_mem(s->data, s->data_sz, &skel_opts);
11049         if (IS_ERR(obj)) {
11050                 pr_warn("failed to initialize skeleton BPF object '%s': %ld\n",
11051                         s->name, PTR_ERR(obj));
11052                 return PTR_ERR(obj);
11053         }
11054
11055         *s->obj = obj;
11056
11057         for (i = 0; i < s->map_cnt; i++) {
11058                 struct bpf_map **map = s->maps[i].map;
11059                 const char *name = s->maps[i].name;
11060                 void **mmaped = s->maps[i].mmaped;
11061
11062                 *map = bpf_object__find_map_by_name(obj, name);
11063                 if (!*map) {
11064                         pr_warn("failed to find skeleton map '%s'\n", name);
11065                         return -ESRCH;
11066                 }
11067
11068                 /* externs shouldn't be pre-setup from user code */
11069                 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
11070                         *mmaped = (*map)->mmaped;
11071         }
11072
11073         for (i = 0; i < s->prog_cnt; i++) {
11074                 struct bpf_program **prog = s->progs[i].prog;
11075                 const char *name = s->progs[i].name;
11076
11077                 *prog = bpf_object__find_program_by_name(obj, name);
11078                 if (!*prog) {
11079                         pr_warn("failed to find skeleton program '%s'\n", name);
11080                         return -ESRCH;
11081                 }
11082         }
11083
11084         return 0;
11085 }
11086
11087 int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
11088 {
11089         int i, err;
11090
11091         err = bpf_object__load(*s->obj);
11092         if (err) {
11093                 pr_warn("failed to load BPF skeleton '%s': %d\n", s->name, err);
11094                 return err;
11095         }
11096
11097         for (i = 0; i < s->map_cnt; i++) {
11098                 struct bpf_map *map = *s->maps[i].map;
11099                 size_t mmap_sz = bpf_map_mmap_sz(map);
11100                 int prot, map_fd = bpf_map__fd(map);
11101                 void **mmaped = s->maps[i].mmaped;
11102
11103                 if (!mmaped)
11104                         continue;
11105
11106                 if (!(map->def.map_flags & BPF_F_MMAPABLE)) {
11107                         *mmaped = NULL;
11108                         continue;
11109                 }
11110
11111                 if (map->def.map_flags & BPF_F_RDONLY_PROG)
11112                         prot = PROT_READ;
11113                 else
11114                         prot = PROT_READ | PROT_WRITE;
11115
11116                 /* Remap anonymous mmap()-ed "map initialization image" as
11117                  * a BPF map-backed mmap()-ed memory, but preserving the same
11118                  * memory address. This will cause kernel to change process'
11119                  * page table to point to a different piece of kernel memory,
11120                  * but from userspace point of view memory address (and its
11121                  * contents, being identical at this point) will stay the
11122                  * same. This mapping will be released by bpf_object__close()
11123                  * as per normal clean up procedure, so we don't need to worry
11124                  * about it from skeleton's clean up perspective.
11125                  */
11126                 *mmaped = mmap(map->mmaped, mmap_sz, prot,
11127                                 MAP_SHARED | MAP_FIXED, map_fd, 0);
11128                 if (*mmaped == MAP_FAILED) {
11129                         err = -errno;
11130                         *mmaped = NULL;
11131                         pr_warn("failed to re-mmap() map '%s': %d\n",
11132                                  bpf_map__name(map), err);
11133                         return err;
11134                 }
11135         }
11136
11137         return 0;
11138 }
11139
11140 int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
11141 {
11142         int i;
11143
11144         for (i = 0; i < s->prog_cnt; i++) {
11145                 struct bpf_program *prog = *s->progs[i].prog;
11146                 struct bpf_link **link = s->progs[i].link;
11147                 const struct bpf_sec_def *sec_def;
11148
11149                 if (!prog->load)
11150                         continue;
11151
11152                 sec_def = find_sec_def(prog->sec_name);
11153                 if (!sec_def || !sec_def->attach_fn)
11154                         continue;
11155
11156                 *link = sec_def->attach_fn(sec_def, prog);
11157                 if (IS_ERR(*link)) {
11158                         pr_warn("failed to auto-attach program '%s': %ld\n",
11159                                 bpf_program__name(prog), PTR_ERR(*link));
11160                         return PTR_ERR(*link);
11161                 }
11162         }
11163
11164         return 0;
11165 }
11166
11167 void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
11168 {
11169         int i;
11170
11171         for (i = 0; i < s->prog_cnt; i++) {
11172                 struct bpf_link **link = s->progs[i].link;
11173
11174                 bpf_link__destroy(*link);
11175                 *link = NULL;
11176         }
11177 }
11178
11179 void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
11180 {
11181         if (s->progs)
11182                 bpf_object__detach_skeleton(s);
11183         if (s->obj)
11184                 bpf_object__close(*s->obj);
11185         free(s->maps);
11186         free(s->progs);
11187         free(s);
11188 }