linux/tools/lib/bpf/btf.c
<<
>>
Prefs
   1// SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
   2/* Copyright (c) 2018 Facebook */
   3
   4#include <byteswap.h>
   5#include <endian.h>
   6#include <stdio.h>
   7#include <stdlib.h>
   8#include <string.h>
   9#include <fcntl.h>
  10#include <unistd.h>
  11#include <errno.h>
  12#include <sys/utsname.h>
  13#include <sys/param.h>
  14#include <sys/stat.h>
  15#include <linux/kernel.h>
  16#include <linux/err.h>
  17#include <linux/btf.h>
  18#include <gelf.h>
  19#include "btf.h"
  20#include "bpf.h"
  21#include "libbpf.h"
  22#include "libbpf_internal.h"
  23#include "hashmap.h"
  24#include "strset.h"
  25
  26#define BTF_MAX_NR_TYPES 0x7fffffffU
  27#define BTF_MAX_STR_OFFSET 0x7fffffffU
  28
  29static struct btf_type btf_void;
  30
  31struct btf {
  32        /* raw BTF data in native endianness */
  33        void *raw_data;
  34        /* raw BTF data in non-native endianness */
  35        void *raw_data_swapped;
  36        __u32 raw_size;
  37        /* whether target endianness differs from the native one */
  38        bool swapped_endian;
  39
  40        /*
  41         * When BTF is loaded from an ELF or raw memory it is stored
  42         * in a contiguous memory block. The hdr, type_data, and, strs_data
  43         * point inside that memory region to their respective parts of BTF
  44         * representation:
  45         *
  46         * +--------------------------------+
  47         * |  Header  |  Types  |  Strings  |
  48         * +--------------------------------+
  49         * ^          ^         ^
  50         * |          |         |
  51         * hdr        |         |
  52         * types_data-+         |
  53         * strs_data------------+
  54         *
  55         * If BTF data is later modified, e.g., due to types added or
  56         * removed, BTF deduplication performed, etc, this contiguous
  57         * representation is broken up into three independently allocated
  58         * memory regions to be able to modify them independently.
  59         * raw_data is nulled out at that point, but can be later allocated
  60         * and cached again if user calls btf__get_raw_data(), at which point
  61         * raw_data will contain a contiguous copy of header, types, and
  62         * strings:
  63         *
  64         * +----------+  +---------+  +-----------+
  65         * |  Header  |  |  Types  |  |  Strings  |
  66         * +----------+  +---------+  +-----------+
  67         * ^             ^            ^
  68         * |             |            |
  69         * hdr           |            |
  70         * types_data----+            |
  71         * strset__data(strs_set)-----+
  72         *
  73         *               +----------+---------+-----------+
  74         *               |  Header  |  Types  |  Strings  |
  75         * raw_data----->+----------+---------+-----------+
  76         */
  77        struct btf_header *hdr;
  78
  79        void *types_data;
  80        size_t types_data_cap; /* used size stored in hdr->type_len */
  81
  82        /* type ID to `struct btf_type *` lookup index
  83         * type_offs[0] corresponds to the first non-VOID type:
  84         *   - for base BTF it's type [1];
  85         *   - for split BTF it's the first non-base BTF type.
  86         */
  87        __u32 *type_offs;
  88        size_t type_offs_cap;
  89        /* number of types in this BTF instance:
  90         *   - doesn't include special [0] void type;
  91         *   - for split BTF counts number of types added on top of base BTF.
  92         */
  93        __u32 nr_types;
  94        /* if not NULL, points to the base BTF on top of which the current
  95         * split BTF is based
  96         */
  97        struct btf *base_btf;
  98        /* BTF type ID of the first type in this BTF instance:
  99         *   - for base BTF it's equal to 1;
 100         *   - for split BTF it's equal to biggest type ID of base BTF plus 1.
 101         */
 102        int start_id;
 103        /* logical string offset of this BTF instance:
 104         *   - for base BTF it's equal to 0;
 105         *   - for split BTF it's equal to total size of base BTF's string section size.
 106         */
 107        int start_str_off;
 108
 109        /* only one of strs_data or strs_set can be non-NULL, depending on
 110         * whether BTF is in a modifiable state (strs_set is used) or not
 111         * (strs_data points inside raw_data)
 112         */
 113        void *strs_data;
 114        /* a set of unique strings */
 115        struct strset *strs_set;
 116        /* whether strings are already deduplicated */
 117        bool strs_deduped;
 118
 119        /* BTF object FD, if loaded into kernel */
 120        int fd;
 121
 122        /* Pointer size (in bytes) for a target architecture of this BTF */
 123        int ptr_sz;
 124};
 125
 126static inline __u64 ptr_to_u64(const void *ptr)
 127{
 128        return (__u64) (unsigned long) ptr;
 129}
 130
 131/* Ensure given dynamically allocated memory region pointed to by *data* with
 132 * capacity of *cap_cnt* elements each taking *elem_sz* bytes has enough
 133 * memory to accomodate *add_cnt* new elements, assuming *cur_cnt* elements
 134 * are already used. At most *max_cnt* elements can be ever allocated.
 135 * If necessary, memory is reallocated and all existing data is copied over,
 136 * new pointer to the memory region is stored at *data, new memory region
 137 * capacity (in number of elements) is stored in *cap.
 138 * On success, memory pointer to the beginning of unused memory is returned.
 139 * On error, NULL is returned.
 140 */
 141void *libbpf_add_mem(void **data, size_t *cap_cnt, size_t elem_sz,
 142                     size_t cur_cnt, size_t max_cnt, size_t add_cnt)
 143{
 144        size_t new_cnt;
 145        void *new_data;
 146
 147        if (cur_cnt + add_cnt <= *cap_cnt)
 148                return *data + cur_cnt * elem_sz;
 149
 150        /* requested more than the set limit */
 151        if (cur_cnt + add_cnt > max_cnt)
 152                return NULL;
 153
 154        new_cnt = *cap_cnt;
 155        new_cnt += new_cnt / 4;           /* expand by 25% */
 156        if (new_cnt < 16)                 /* but at least 16 elements */
 157                new_cnt = 16;
 158        if (new_cnt > max_cnt)            /* but not exceeding a set limit */
 159                new_cnt = max_cnt;
 160        if (new_cnt < cur_cnt + add_cnt)  /* also ensure we have enough memory */
 161                new_cnt = cur_cnt + add_cnt;
 162
 163        new_data = libbpf_reallocarray(*data, new_cnt, elem_sz);
 164        if (!new_data)
 165                return NULL;
 166
 167        /* zero out newly allocated portion of memory */
 168        memset(new_data + (*cap_cnt) * elem_sz, 0, (new_cnt - *cap_cnt) * elem_sz);
 169
 170        *data = new_data;
 171        *cap_cnt = new_cnt;
 172        return new_data + cur_cnt * elem_sz;
 173}
 174
 175/* Ensure given dynamically allocated memory region has enough allocated space
 176 * to accommodate *need_cnt* elements of size *elem_sz* bytes each
 177 */
 178int libbpf_ensure_mem(void **data, size_t *cap_cnt, size_t elem_sz, size_t need_cnt)
 179{
 180        void *p;
 181
 182        if (need_cnt <= *cap_cnt)
 183                return 0;
 184
 185        p = libbpf_add_mem(data, cap_cnt, elem_sz, *cap_cnt, SIZE_MAX, need_cnt - *cap_cnt);
 186        if (!p)
 187                return -ENOMEM;
 188
 189        return 0;
 190}
 191
 192static int btf_add_type_idx_entry(struct btf *btf, __u32 type_off)
 193{
 194        __u32 *p;
 195
 196        p = libbpf_add_mem((void **)&btf->type_offs, &btf->type_offs_cap, sizeof(__u32),
 197                           btf->nr_types, BTF_MAX_NR_TYPES, 1);
 198        if (!p)
 199                return -ENOMEM;
 200
 201        *p = type_off;
 202        return 0;
 203}
 204
 205static void btf_bswap_hdr(struct btf_header *h)
 206{
 207        h->magic = bswap_16(h->magic);
 208        h->hdr_len = bswap_32(h->hdr_len);
 209        h->type_off = bswap_32(h->type_off);
 210        h->type_len = bswap_32(h->type_len);
 211        h->str_off = bswap_32(h->str_off);
 212        h->str_len = bswap_32(h->str_len);
 213}
 214
 215static int btf_parse_hdr(struct btf *btf)
 216{
 217        struct btf_header *hdr = btf->hdr;
 218        __u32 meta_left;
 219
 220        if (btf->raw_size < sizeof(struct btf_header)) {
 221                pr_debug("BTF header not found\n");
 222                return -EINVAL;
 223        }
 224
 225        if (hdr->magic == bswap_16(BTF_MAGIC)) {
 226                btf->swapped_endian = true;
 227                if (bswap_32(hdr->hdr_len) != sizeof(struct btf_header)) {
 228                        pr_warn("Can't load BTF with non-native endianness due to unsupported header length %u\n",
 229                                bswap_32(hdr->hdr_len));
 230                        return -ENOTSUP;
 231                }
 232                btf_bswap_hdr(hdr);
 233        } else if (hdr->magic != BTF_MAGIC) {
 234                pr_debug("Invalid BTF magic:%x\n", hdr->magic);
 235                return -EINVAL;
 236        }
 237
 238        meta_left = btf->raw_size - sizeof(*hdr);
 239        if (meta_left < hdr->str_off + hdr->str_len) {
 240                pr_debug("Invalid BTF total size:%u\n", btf->raw_size);
 241                return -EINVAL;
 242        }
 243
 244        if (hdr->type_off + hdr->type_len > hdr->str_off) {
 245                pr_debug("Invalid BTF data sections layout: type data at %u + %u, strings data at %u + %u\n",
 246                         hdr->type_off, hdr->type_len, hdr->str_off, hdr->str_len);
 247                return -EINVAL;
 248        }
 249
 250        if (hdr->type_off % 4) {
 251                pr_debug("BTF type section is not aligned to 4 bytes\n");
 252                return -EINVAL;
 253        }
 254
 255        return 0;
 256}
 257
 258static int btf_parse_str_sec(struct btf *btf)
 259{
 260        const struct btf_header *hdr = btf->hdr;
 261        const char *start = btf->strs_data;
 262        const char *end = start + btf->hdr->str_len;
 263
 264        if (btf->base_btf && hdr->str_len == 0)
 265                return 0;
 266        if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET || end[-1]) {
 267                pr_debug("Invalid BTF string section\n");
 268                return -EINVAL;
 269        }
 270        if (!btf->base_btf && start[0]) {
 271                pr_debug("Invalid BTF string section\n");
 272                return -EINVAL;
 273        }
 274        return 0;
 275}
 276
 277static int btf_type_size(const struct btf_type *t)
 278{
 279        const int base_size = sizeof(struct btf_type);
 280        __u16 vlen = btf_vlen(t);
 281
 282        switch (btf_kind(t)) {
 283        case BTF_KIND_FWD:
 284        case BTF_KIND_CONST:
 285        case BTF_KIND_VOLATILE:
 286        case BTF_KIND_RESTRICT:
 287        case BTF_KIND_PTR:
 288        case BTF_KIND_TYPEDEF:
 289        case BTF_KIND_FUNC:
 290        case BTF_KIND_FLOAT:
 291                return base_size;
 292        case BTF_KIND_INT:
 293                return base_size + sizeof(__u32);
 294        case BTF_KIND_ENUM:
 295                return base_size + vlen * sizeof(struct btf_enum);
 296        case BTF_KIND_ARRAY:
 297                return base_size + sizeof(struct btf_array);
 298        case BTF_KIND_STRUCT:
 299        case BTF_KIND_UNION:
 300                return base_size + vlen * sizeof(struct btf_member);
 301        case BTF_KIND_FUNC_PROTO:
 302                return base_size + vlen * sizeof(struct btf_param);
 303        case BTF_KIND_VAR:
 304                return base_size + sizeof(struct btf_var);
 305        case BTF_KIND_DATASEC:
 306                return base_size + vlen * sizeof(struct btf_var_secinfo);
 307        default:
 308                pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
 309                return -EINVAL;
 310        }
 311}
 312
 313static void btf_bswap_type_base(struct btf_type *t)
 314{
 315        t->name_off = bswap_32(t->name_off);
 316        t->info = bswap_32(t->info);
 317        t->type = bswap_32(t->type);
 318}
 319
 320static int btf_bswap_type_rest(struct btf_type *t)
 321{
 322        struct btf_var_secinfo *v;
 323        struct btf_member *m;
 324        struct btf_array *a;
 325        struct btf_param *p;
 326        struct btf_enum *e;
 327        __u16 vlen = btf_vlen(t);
 328        int i;
 329
 330        switch (btf_kind(t)) {
 331        case BTF_KIND_FWD:
 332        case BTF_KIND_CONST:
 333        case BTF_KIND_VOLATILE:
 334        case BTF_KIND_RESTRICT:
 335        case BTF_KIND_PTR:
 336        case BTF_KIND_TYPEDEF:
 337        case BTF_KIND_FUNC:
 338        case BTF_KIND_FLOAT:
 339                return 0;
 340        case BTF_KIND_INT:
 341                *(__u32 *)(t + 1) = bswap_32(*(__u32 *)(t + 1));
 342                return 0;
 343        case BTF_KIND_ENUM:
 344                for (i = 0, e = btf_enum(t); i < vlen; i++, e++) {
 345                        e->name_off = bswap_32(e->name_off);
 346                        e->val = bswap_32(e->val);
 347                }
 348                return 0;
 349        case BTF_KIND_ARRAY:
 350                a = btf_array(t);
 351                a->type = bswap_32(a->type);
 352                a->index_type = bswap_32(a->index_type);
 353                a->nelems = bswap_32(a->nelems);
 354                return 0;
 355        case BTF_KIND_STRUCT:
 356        case BTF_KIND_UNION:
 357                for (i = 0, m = btf_members(t); i < vlen; i++, m++) {
 358                        m->name_off = bswap_32(m->name_off);
 359                        m->type = bswap_32(m->type);
 360                        m->offset = bswap_32(m->offset);
 361                }
 362                return 0;
 363        case BTF_KIND_FUNC_PROTO:
 364                for (i = 0, p = btf_params(t); i < vlen; i++, p++) {
 365                        p->name_off = bswap_32(p->name_off);
 366                        p->type = bswap_32(p->type);
 367                }
 368                return 0;
 369        case BTF_KIND_VAR:
 370                btf_var(t)->linkage = bswap_32(btf_var(t)->linkage);
 371                return 0;
 372        case BTF_KIND_DATASEC:
 373                for (i = 0, v = btf_var_secinfos(t); i < vlen; i++, v++) {
 374                        v->type = bswap_32(v->type);
 375                        v->offset = bswap_32(v->offset);
 376                        v->size = bswap_32(v->size);
 377                }
 378                return 0;
 379        default:
 380                pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
 381                return -EINVAL;
 382        }
 383}
 384
 385static int btf_parse_type_sec(struct btf *btf)
 386{
 387        struct btf_header *hdr = btf->hdr;
 388        void *next_type = btf->types_data;
 389        void *end_type = next_type + hdr->type_len;
 390        int err, type_size;
 391
 392        while (next_type + sizeof(struct btf_type) <= end_type) {
 393                if (btf->swapped_endian)
 394                        btf_bswap_type_base(next_type);
 395
 396                type_size = btf_type_size(next_type);
 397                if (type_size < 0)
 398                        return type_size;
 399                if (next_type + type_size > end_type) {
 400                        pr_warn("BTF type [%d] is malformed\n", btf->start_id + btf->nr_types);
 401                        return -EINVAL;
 402                }
 403
 404                if (btf->swapped_endian && btf_bswap_type_rest(next_type))
 405                        return -EINVAL;
 406
 407                err = btf_add_type_idx_entry(btf, next_type - btf->types_data);
 408                if (err)
 409                        return err;
 410
 411                next_type += type_size;
 412                btf->nr_types++;
 413        }
 414
 415        if (next_type != end_type) {
 416                pr_warn("BTF types data is malformed\n");
 417                return -EINVAL;
 418        }
 419
 420        return 0;
 421}
 422
 423__u32 btf__get_nr_types(const struct btf *btf)
 424{
 425        return btf->start_id + btf->nr_types - 1;
 426}
 427
 428const struct btf *btf__base_btf(const struct btf *btf)
 429{
 430        return btf->base_btf;
 431}
 432
 433/* internal helper returning non-const pointer to a type */
 434struct btf_type *btf_type_by_id(struct btf *btf, __u32 type_id)
 435{
 436        if (type_id == 0)
 437                return &btf_void;
 438        if (type_id < btf->start_id)
 439                return btf_type_by_id(btf->base_btf, type_id);
 440        return btf->types_data + btf->type_offs[type_id - btf->start_id];
 441}
 442
 443const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
 444{
 445        if (type_id >= btf->start_id + btf->nr_types)
 446                return errno = EINVAL, NULL;
 447        return btf_type_by_id((struct btf *)btf, type_id);
 448}
 449
 450static int determine_ptr_size(const struct btf *btf)
 451{
 452        const struct btf_type *t;
 453        const char *name;
 454        int i, n;
 455
 456        if (btf->base_btf && btf->base_btf->ptr_sz > 0)
 457                return btf->base_btf->ptr_sz;
 458
 459        n = btf__get_nr_types(btf);
 460        for (i = 1; i <= n; i++) {
 461                t = btf__type_by_id(btf, i);
 462                if (!btf_is_int(t))
 463                        continue;
 464
 465                name = btf__name_by_offset(btf, t->name_off);
 466                if (!name)
 467                        continue;
 468
 469                if (strcmp(name, "long int") == 0 ||
 470                    strcmp(name, "long unsigned int") == 0) {
 471                        if (t->size != 4 && t->size != 8)
 472                                continue;
 473                        return t->size;
 474                }
 475        }
 476
 477        return -1;
 478}
 479
 480static size_t btf_ptr_sz(const struct btf *btf)
 481{
 482        if (!btf->ptr_sz)
 483                ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
 484        return btf->ptr_sz < 0 ? sizeof(void *) : btf->ptr_sz;
 485}
 486
 487/* Return pointer size this BTF instance assumes. The size is heuristically
 488 * determined by looking for 'long' or 'unsigned long' integer type and
 489 * recording its size in bytes. If BTF type information doesn't have any such
 490 * type, this function returns 0. In the latter case, native architecture's
 491 * pointer size is assumed, so will be either 4 or 8, depending on
 492 * architecture that libbpf was compiled for. It's possible to override
 493 * guessed value by using btf__set_pointer_size() API.
 494 */
 495size_t btf__pointer_size(const struct btf *btf)
 496{
 497        if (!btf->ptr_sz)
 498                ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
 499
 500        if (btf->ptr_sz < 0)
 501                /* not enough BTF type info to guess */
 502                return 0;
 503
 504        return btf->ptr_sz;
 505}
 506
 507/* Override or set pointer size in bytes. Only values of 4 and 8 are
 508 * supported.
 509 */
 510int btf__set_pointer_size(struct btf *btf, size_t ptr_sz)
 511{
 512        if (ptr_sz != 4 && ptr_sz != 8)
 513                return libbpf_err(-EINVAL);
 514        btf->ptr_sz = ptr_sz;
 515        return 0;
 516}
 517
 518static bool is_host_big_endian(void)
 519{
 520#if __BYTE_ORDER == __LITTLE_ENDIAN
 521        return false;
 522#elif __BYTE_ORDER == __BIG_ENDIAN
 523        return true;
 524#else
 525# error "Unrecognized __BYTE_ORDER__"
 526#endif
 527}
 528
 529enum btf_endianness btf__endianness(const struct btf *btf)
 530{
 531        if (is_host_big_endian())
 532                return btf->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
 533        else
 534                return btf->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
 535}
 536
 537int btf__set_endianness(struct btf *btf, enum btf_endianness endian)
 538{
 539        if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
 540                return libbpf_err(-EINVAL);
 541
 542        btf->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
 543        if (!btf->swapped_endian) {
 544                free(btf->raw_data_swapped);
 545                btf->raw_data_swapped = NULL;
 546        }
 547        return 0;
 548}
 549
 550static bool btf_type_is_void(const struct btf_type *t)
 551{
 552        return t == &btf_void || btf_is_fwd(t);
 553}
 554
 555static bool btf_type_is_void_or_null(const struct btf_type *t)
 556{
 557        return !t || btf_type_is_void(t);
 558}
 559
 560#define MAX_RESOLVE_DEPTH 32
 561
 562__s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
 563{
 564        const struct btf_array *array;
 565        const struct btf_type *t;
 566        __u32 nelems = 1;
 567        __s64 size = -1;
 568        int i;
 569
 570        t = btf__type_by_id(btf, type_id);
 571        for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t); i++) {
 572                switch (btf_kind(t)) {
 573                case BTF_KIND_INT:
 574                case BTF_KIND_STRUCT:
 575                case BTF_KIND_UNION:
 576                case BTF_KIND_ENUM:
 577                case BTF_KIND_DATASEC:
 578                case BTF_KIND_FLOAT:
 579                        size = t->size;
 580                        goto done;
 581                case BTF_KIND_PTR:
 582                        size = btf_ptr_sz(btf);
 583                        goto done;
 584                case BTF_KIND_TYPEDEF:
 585                case BTF_KIND_VOLATILE:
 586                case BTF_KIND_CONST:
 587                case BTF_KIND_RESTRICT:
 588                case BTF_KIND_VAR:
 589                        type_id = t->type;
 590                        break;
 591                case BTF_KIND_ARRAY:
 592                        array = btf_array(t);
 593                        if (nelems && array->nelems > UINT32_MAX / nelems)
 594                                return libbpf_err(-E2BIG);
 595                        nelems *= array->nelems;
 596                        type_id = array->type;
 597                        break;
 598                default:
 599                        return libbpf_err(-EINVAL);
 600                }
 601
 602                t = btf__type_by_id(btf, type_id);
 603        }
 604
 605done:
 606        if (size < 0)
 607                return libbpf_err(-EINVAL);
 608        if (nelems && size > UINT32_MAX / nelems)
 609                return libbpf_err(-E2BIG);
 610
 611        return nelems * size;
 612}
 613
 614int btf__align_of(const struct btf *btf, __u32 id)
 615{
 616        const struct btf_type *t = btf__type_by_id(btf, id);
 617        __u16 kind = btf_kind(t);
 618
 619        switch (kind) {
 620        case BTF_KIND_INT:
 621        case BTF_KIND_ENUM:
 622        case BTF_KIND_FLOAT:
 623                return min(btf_ptr_sz(btf), (size_t)t->size);
 624        case BTF_KIND_PTR:
 625                return btf_ptr_sz(btf);
 626        case BTF_KIND_TYPEDEF:
 627        case BTF_KIND_VOLATILE:
 628        case BTF_KIND_CONST:
 629        case BTF_KIND_RESTRICT:
 630                return btf__align_of(btf, t->type);
 631        case BTF_KIND_ARRAY:
 632                return btf__align_of(btf, btf_array(t)->type);
 633        case BTF_KIND_STRUCT:
 634        case BTF_KIND_UNION: {
 635                const struct btf_member *m = btf_members(t);
 636                __u16 vlen = btf_vlen(t);
 637                int i, max_align = 1, align;
 638
 639                for (i = 0; i < vlen; i++, m++) {
 640                        align = btf__align_of(btf, m->type);
 641                        if (align <= 0)
 642                                return libbpf_err(align);
 643                        max_align = max(max_align, align);
 644                }
 645
 646                return max_align;
 647        }
 648        default:
 649                pr_warn("unsupported BTF_KIND:%u\n", btf_kind(t));
 650                return errno = EINVAL, 0;
 651        }
 652}
 653
 654int btf__resolve_type(const struct btf *btf, __u32 type_id)
 655{
 656        const struct btf_type *t;
 657        int depth = 0;
 658
 659        t = btf__type_by_id(btf, type_id);
 660        while (depth < MAX_RESOLVE_DEPTH &&
 661               !btf_type_is_void_or_null(t) &&
 662               (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
 663                type_id = t->type;
 664                t = btf__type_by_id(btf, type_id);
 665                depth++;
 666        }
 667
 668        if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
 669                return libbpf_err(-EINVAL);
 670
 671        return type_id;
 672}
 673
 674__s32 btf__find_by_name(const struct btf *btf, const char *type_name)
 675{
 676        __u32 i, nr_types = btf__get_nr_types(btf);
 677
 678        if (!strcmp(type_name, "void"))
 679                return 0;
 680
 681        for (i = 1; i <= nr_types; i++) {
 682                const struct btf_type *t = btf__type_by_id(btf, i);
 683                const char *name = btf__name_by_offset(btf, t->name_off);
 684
 685                if (name && !strcmp(type_name, name))
 686                        return i;
 687        }
 688
 689        return libbpf_err(-ENOENT);
 690}
 691
 692__s32 btf__find_by_name_kind(const struct btf *btf, const char *type_name,
 693                             __u32 kind)
 694{
 695        __u32 i, nr_types = btf__get_nr_types(btf);
 696
 697        if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
 698                return 0;
 699
 700        for (i = 1; i <= nr_types; i++) {
 701                const struct btf_type *t = btf__type_by_id(btf, i);
 702                const char *name;
 703
 704                if (btf_kind(t) != kind)
 705                        continue;
 706                name = btf__name_by_offset(btf, t->name_off);
 707                if (name && !strcmp(type_name, name))
 708                        return i;
 709        }
 710
 711        return libbpf_err(-ENOENT);
 712}
 713
 714static bool btf_is_modifiable(const struct btf *btf)
 715{
 716        return (void *)btf->hdr != btf->raw_data;
 717}
 718
 719void btf__free(struct btf *btf)
 720{
 721        if (IS_ERR_OR_NULL(btf))
 722                return;
 723
 724        if (btf->fd >= 0)
 725                close(btf->fd);
 726
 727        if (btf_is_modifiable(btf)) {
 728                /* if BTF was modified after loading, it will have a split
 729                 * in-memory representation for header, types, and strings
 730                 * sections, so we need to free all of them individually. It
 731                 * might still have a cached contiguous raw data present,
 732                 * which will be unconditionally freed below.
 733                 */
 734                free(btf->hdr);
 735                free(btf->types_data);
 736                strset__free(btf->strs_set);
 737        }
 738        free(btf->raw_data);
 739        free(btf->raw_data_swapped);
 740        free(btf->type_offs);
 741        free(btf);
 742}
 743
 744static struct btf *btf_new_empty(struct btf *base_btf)
 745{
 746        struct btf *btf;
 747
 748        btf = calloc(1, sizeof(*btf));
 749        if (!btf)
 750                return ERR_PTR(-ENOMEM);
 751
 752        btf->nr_types = 0;
 753        btf->start_id = 1;
 754        btf->start_str_off = 0;
 755        btf->fd = -1;
 756        btf->ptr_sz = sizeof(void *);
 757        btf->swapped_endian = false;
 758
 759        if (base_btf) {
 760                btf->base_btf = base_btf;
 761                btf->start_id = btf__get_nr_types(base_btf) + 1;
 762                btf->start_str_off = base_btf->hdr->str_len;
 763        }
 764
 765        /* +1 for empty string at offset 0 */
 766        btf->raw_size = sizeof(struct btf_header) + (base_btf ? 0 : 1);
 767        btf->raw_data = calloc(1, btf->raw_size);
 768        if (!btf->raw_data) {
 769                free(btf);
 770                return ERR_PTR(-ENOMEM);
 771        }
 772
 773        btf->hdr = btf->raw_data;
 774        btf->hdr->hdr_len = sizeof(struct btf_header);
 775        btf->hdr->magic = BTF_MAGIC;
 776        btf->hdr->version = BTF_VERSION;
 777
 778        btf->types_data = btf->raw_data + btf->hdr->hdr_len;
 779        btf->strs_data = btf->raw_data + btf->hdr->hdr_len;
 780        btf->hdr->str_len = base_btf ? 0 : 1; /* empty string at offset 0 */
 781
 782        return btf;
 783}
 784
 785struct btf *btf__new_empty(void)
 786{
 787        return libbpf_ptr(btf_new_empty(NULL));
 788}
 789
 790struct btf *btf__new_empty_split(struct btf *base_btf)
 791{
 792        return libbpf_ptr(btf_new_empty(base_btf));
 793}
 794
 795static struct btf *btf_new(const void *data, __u32 size, struct btf *base_btf)
 796{
 797        struct btf *btf;
 798        int err;
 799
 800        btf = calloc(1, sizeof(struct btf));
 801        if (!btf)
 802                return ERR_PTR(-ENOMEM);
 803
 804        btf->nr_types = 0;
 805        btf->start_id = 1;
 806        btf->start_str_off = 0;
 807        btf->fd = -1;
 808
 809        if (base_btf) {
 810                btf->base_btf = base_btf;
 811                btf->start_id = btf__get_nr_types(base_btf) + 1;
 812                btf->start_str_off = base_btf->hdr->str_len;
 813        }
 814
 815        btf->raw_data = malloc(size);
 816        if (!btf->raw_data) {
 817                err = -ENOMEM;
 818                goto done;
 819        }
 820        memcpy(btf->raw_data, data, size);
 821        btf->raw_size = size;
 822
 823        btf->hdr = btf->raw_data;
 824        err = btf_parse_hdr(btf);
 825        if (err)
 826                goto done;
 827
 828        btf->strs_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->str_off;
 829        btf->types_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->type_off;
 830
 831        err = btf_parse_str_sec(btf);
 832        err = err ?: btf_parse_type_sec(btf);
 833        if (err)
 834                goto done;
 835
 836done:
 837        if (err) {
 838                btf__free(btf);
 839                return ERR_PTR(err);
 840        }
 841
 842        return btf;
 843}
 844
 845struct btf *btf__new(const void *data, __u32 size)
 846{
 847        return libbpf_ptr(btf_new(data, size, NULL));
 848}
 849
 850static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,
 851                                 struct btf_ext **btf_ext)
 852{
 853        Elf_Data *btf_data = NULL, *btf_ext_data = NULL;
 854        int err = 0, fd = -1, idx = 0;
 855        struct btf *btf = NULL;
 856        Elf_Scn *scn = NULL;
 857        Elf *elf = NULL;
 858        GElf_Ehdr ehdr;
 859        size_t shstrndx;
 860
 861        if (elf_version(EV_CURRENT) == EV_NONE) {
 862                pr_warn("failed to init libelf for %s\n", path);
 863                return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
 864        }
 865
 866        fd = open(path, O_RDONLY);
 867        if (fd < 0) {
 868                err = -errno;
 869                pr_warn("failed to open %s: %s\n", path, strerror(errno));
 870                return ERR_PTR(err);
 871        }
 872
 873        err = -LIBBPF_ERRNO__FORMAT;
 874
 875        elf = elf_begin(fd, ELF_C_READ, NULL);
 876        if (!elf) {
 877                pr_warn("failed to open %s as ELF file\n", path);
 878                goto done;
 879        }
 880        if (!gelf_getehdr(elf, &ehdr)) {
 881                pr_warn("failed to get EHDR from %s\n", path);
 882                goto done;
 883        }
 884
 885        if (elf_getshdrstrndx(elf, &shstrndx)) {
 886                pr_warn("failed to get section names section index for %s\n",
 887                        path);
 888                goto done;
 889        }
 890
 891        if (!elf_rawdata(elf_getscn(elf, shstrndx), NULL)) {
 892                pr_warn("failed to get e_shstrndx from %s\n", path);
 893                goto done;
 894        }
 895
 896        while ((scn = elf_nextscn(elf, scn)) != NULL) {
 897                GElf_Shdr sh;
 898                char *name;
 899
 900                idx++;
 901                if (gelf_getshdr(scn, &sh) != &sh) {
 902                        pr_warn("failed to get section(%d) header from %s\n",
 903                                idx, path);
 904                        goto done;
 905                }
 906                name = elf_strptr(elf, shstrndx, sh.sh_name);
 907                if (!name) {
 908                        pr_warn("failed to get section(%d) name from %s\n",
 909                                idx, path);
 910                        goto done;
 911                }
 912                if (strcmp(name, BTF_ELF_SEC) == 0) {
 913                        btf_data = elf_getdata(scn, 0);
 914                        if (!btf_data) {
 915                                pr_warn("failed to get section(%d, %s) data from %s\n",
 916                                        idx, name, path);
 917                                goto done;
 918                        }
 919                        continue;
 920                } else if (btf_ext && strcmp(name, BTF_EXT_ELF_SEC) == 0) {
 921                        btf_ext_data = elf_getdata(scn, 0);
 922                        if (!btf_ext_data) {
 923                                pr_warn("failed to get section(%d, %s) data from %s\n",
 924                                        idx, name, path);
 925                                goto done;
 926                        }
 927                        continue;
 928                }
 929        }
 930
 931        err = 0;
 932
 933        if (!btf_data) {
 934                err = -ENOENT;
 935                goto done;
 936        }
 937        btf = btf_new(btf_data->d_buf, btf_data->d_size, base_btf);
 938        err = libbpf_get_error(btf);
 939        if (err)
 940                goto done;
 941
 942        switch (gelf_getclass(elf)) {
 943        case ELFCLASS32:
 944                btf__set_pointer_size(btf, 4);
 945                break;
 946        case ELFCLASS64:
 947                btf__set_pointer_size(btf, 8);
 948                break;
 949        default:
 950                pr_warn("failed to get ELF class (bitness) for %s\n", path);
 951                break;
 952        }
 953
 954        if (btf_ext && btf_ext_data) {
 955                *btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
 956                err = libbpf_get_error(*btf_ext);
 957                if (err)
 958                        goto done;
 959        } else if (btf_ext) {
 960                *btf_ext = NULL;
 961        }
 962done:
 963        if (elf)
 964                elf_end(elf);
 965        close(fd);
 966
 967        if (!err)
 968                return btf;
 969
 970        if (btf_ext)
 971                btf_ext__free(*btf_ext);
 972        btf__free(btf);
 973
 974        return ERR_PTR(err);
 975}
 976
 977struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
 978{
 979        return libbpf_ptr(btf_parse_elf(path, NULL, btf_ext));
 980}
 981
 982struct btf *btf__parse_elf_split(const char *path, struct btf *base_btf)
 983{
 984        return libbpf_ptr(btf_parse_elf(path, base_btf, NULL));
 985}
 986
 987static struct btf *btf_parse_raw(const char *path, struct btf *base_btf)
 988{
 989        struct btf *btf = NULL;
 990        void *data = NULL;
 991        FILE *f = NULL;
 992        __u16 magic;
 993        int err = 0;
 994        long sz;
 995
 996        f = fopen(path, "rb");
 997        if (!f) {
 998                err = -errno;
 999                goto err_out;
1000        }
1001
1002        /* check BTF magic */
1003        if (fread(&magic, 1, sizeof(magic), f) < sizeof(magic)) {
1004                err = -EIO;
1005                goto err_out;
1006        }
1007        if (magic != BTF_MAGIC && magic != bswap_16(BTF_MAGIC)) {
1008                /* definitely not a raw BTF */
1009                err = -EPROTO;
1010                goto err_out;
1011        }
1012
1013        /* get file size */
1014        if (fseek(f, 0, SEEK_END)) {
1015                err = -errno;
1016                goto err_out;
1017        }
1018        sz = ftell(f);
1019        if (sz < 0) {
1020                err = -errno;
1021                goto err_out;
1022        }
1023        /* rewind to the start */
1024        if (fseek(f, 0, SEEK_SET)) {
1025                err = -errno;
1026                goto err_out;
1027        }
1028
1029        /* pre-alloc memory and read all of BTF data */
1030        data = malloc(sz);
1031        if (!data) {
1032                err = -ENOMEM;
1033                goto err_out;
1034        }
1035        if (fread(data, 1, sz, f) < sz) {
1036                err = -EIO;
1037                goto err_out;
1038        }
1039
1040        /* finally parse BTF data */
1041        btf = btf_new(data, sz, base_btf);
1042
1043err_out:
1044        free(data);
1045        if (f)
1046                fclose(f);
1047        return err ? ERR_PTR(err) : btf;
1048}
1049
1050struct btf *btf__parse_raw(const char *path)
1051{
1052        return libbpf_ptr(btf_parse_raw(path, NULL));
1053}
1054
1055struct btf *btf__parse_raw_split(const char *path, struct btf *base_btf)
1056{
1057        return libbpf_ptr(btf_parse_raw(path, base_btf));
1058}
1059
1060static struct btf *btf_parse(const char *path, struct btf *base_btf, struct btf_ext **btf_ext)
1061{
1062        struct btf *btf;
1063        int err;
1064
1065        if (btf_ext)
1066                *btf_ext = NULL;
1067
1068        btf = btf_parse_raw(path, base_btf);
1069        err = libbpf_get_error(btf);
1070        if (!err)
1071                return btf;
1072        if (err != -EPROTO)
1073                return ERR_PTR(err);
1074        return btf_parse_elf(path, base_btf, btf_ext);
1075}
1076
1077struct btf *btf__parse(const char *path, struct btf_ext **btf_ext)
1078{
1079        return libbpf_ptr(btf_parse(path, NULL, btf_ext));
1080}
1081
1082struct btf *btf__parse_split(const char *path, struct btf *base_btf)
1083{
1084        return libbpf_ptr(btf_parse(path, base_btf, NULL));
1085}
1086
1087static int compare_vsi_off(const void *_a, const void *_b)
1088{
1089        const struct btf_var_secinfo *a = _a;
1090        const struct btf_var_secinfo *b = _b;
1091
1092        return a->offset - b->offset;
1093}
1094
1095static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
1096                             struct btf_type *t)
1097{
1098        __u32 size = 0, off = 0, i, vars = btf_vlen(t);
1099        const char *name = btf__name_by_offset(btf, t->name_off);
1100        const struct btf_type *t_var;
1101        struct btf_var_secinfo *vsi;
1102        const struct btf_var *var;
1103        int ret;
1104
1105        if (!name) {
1106                pr_debug("No name found in string section for DATASEC kind.\n");
1107                return -ENOENT;
1108        }
1109
1110        /* .extern datasec size and var offsets were set correctly during
1111         * extern collection step, so just skip straight to sorting variables
1112         */
1113        if (t->size)
1114                goto sort_vars;
1115
1116        ret = bpf_object__section_size(obj, name, &size);
1117        if (ret || !size || (t->size && t->size != size)) {
1118                pr_debug("Invalid size for section %s: %u bytes\n", name, size);
1119                return -ENOENT;
1120        }
1121
1122        t->size = size;
1123
1124        for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
1125                t_var = btf__type_by_id(btf, vsi->type);
1126                var = btf_var(t_var);
1127
1128                if (!btf_is_var(t_var)) {
1129                        pr_debug("Non-VAR type seen in section %s\n", name);
1130                        return -EINVAL;
1131                }
1132
1133                if (var->linkage == BTF_VAR_STATIC)
1134                        continue;
1135
1136                name = btf__name_by_offset(btf, t_var->name_off);
1137                if (!name) {
1138                        pr_debug("No name found in string section for VAR kind\n");
1139                        return -ENOENT;
1140                }
1141
1142                ret = bpf_object__variable_offset(obj, name, &off);
1143                if (ret) {
1144                        pr_debug("No offset found in symbol table for VAR %s\n",
1145                                 name);
1146                        return -ENOENT;
1147                }
1148
1149                vsi->offset = off;
1150        }
1151
1152sort_vars:
1153        qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
1154        return 0;
1155}
1156
1157int btf__finalize_data(struct bpf_object *obj, struct btf *btf)
1158{
1159        int err = 0;
1160        __u32 i;
1161
1162        for (i = 1; i <= btf->nr_types; i++) {
1163                struct btf_type *t = btf_type_by_id(btf, i);
1164
1165                /* Loader needs to fix up some of the things compiler
1166                 * couldn't get its hands on while emitting BTF. This
1167                 * is section size and global variable offset. We use
1168                 * the info from the ELF itself for this purpose.
1169                 */
1170                if (btf_is_datasec(t)) {
1171                        err = btf_fixup_datasec(obj, btf, t);
1172                        if (err)
1173                                break;
1174                }
1175        }
1176
1177        return libbpf_err(err);
1178}
1179
1180static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian);
1181
1182int btf__load_into_kernel(struct btf *btf)
1183{
1184        __u32 log_buf_size = 0, raw_size;
1185        char *log_buf = NULL;
1186        void *raw_data;
1187        int err = 0;
1188
1189        if (btf->fd >= 0)
1190                return libbpf_err(-EEXIST);
1191
1192retry_load:
1193        if (log_buf_size) {
1194                log_buf = malloc(log_buf_size);
1195                if (!log_buf)
1196                        return libbpf_err(-ENOMEM);
1197
1198                *log_buf = 0;
1199        }
1200
1201        raw_data = btf_get_raw_data(btf, &raw_size, false);
1202        if (!raw_data) {
1203                err = -ENOMEM;
1204                goto done;
1205        }
1206        /* cache native raw data representation */
1207        btf->raw_size = raw_size;
1208        btf->raw_data = raw_data;
1209
1210        btf->fd = bpf_load_btf(raw_data, raw_size, log_buf, log_buf_size, false);
1211        if (btf->fd < 0) {
1212                if (!log_buf || errno == ENOSPC) {
1213                        log_buf_size = max((__u32)BPF_LOG_BUF_SIZE,
1214                                           log_buf_size << 1);
1215                        free(log_buf);
1216                        goto retry_load;
1217                }
1218
1219                err = -errno;
1220                pr_warn("Error loading BTF: %s(%d)\n", strerror(errno), errno);
1221                if (*log_buf)
1222                        pr_warn("%s\n", log_buf);
1223                goto done;
1224        }
1225
1226done:
1227        free(log_buf);
1228        return libbpf_err(err);
1229}
1230int btf__load(struct btf *) __attribute__((alias("btf__load_into_kernel")));
1231
1232int btf__fd(const struct btf *btf)
1233{
1234        return btf->fd;
1235}
1236
1237void btf__set_fd(struct btf *btf, int fd)
1238{
1239        btf->fd = fd;
1240}
1241
1242static const void *btf_strs_data(const struct btf *btf)
1243{
1244        return btf->strs_data ? btf->strs_data : strset__data(btf->strs_set);
1245}
1246
1247static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian)
1248{
1249        struct btf_header *hdr = btf->hdr;
1250        struct btf_type *t;
1251        void *data, *p;
1252        __u32 data_sz;
1253        int i;
1254
1255        data = swap_endian ? btf->raw_data_swapped : btf->raw_data;
1256        if (data) {
1257                *size = btf->raw_size;
1258                return data;
1259        }
1260
1261        data_sz = hdr->hdr_len + hdr->type_len + hdr->str_len;
1262        data = calloc(1, data_sz);
1263        if (!data)
1264                return NULL;
1265        p = data;
1266
1267        memcpy(p, hdr, hdr->hdr_len);
1268        if (swap_endian)
1269                btf_bswap_hdr(p);
1270        p += hdr->hdr_len;
1271
1272        memcpy(p, btf->types_data, hdr->type_len);
1273        if (swap_endian) {
1274                for (i = 0; i < btf->nr_types; i++) {
1275                        t = p + btf->type_offs[i];
1276                        /* btf_bswap_type_rest() relies on native t->info, so
1277                         * we swap base type info after we swapped all the
1278                         * additional information
1279                         */
1280                        if (btf_bswap_type_rest(t))
1281                                goto err_out;
1282                        btf_bswap_type_base(t);
1283                }
1284        }
1285        p += hdr->type_len;
1286
1287        memcpy(p, btf_strs_data(btf), hdr->str_len);
1288        p += hdr->str_len;
1289
1290        *size = data_sz;
1291        return data;
1292err_out:
1293        free(data);
1294        return NULL;
1295}
1296
1297const void *btf__get_raw_data(const struct btf *btf_ro, __u32 *size)
1298{
1299        struct btf *btf = (struct btf *)btf_ro;
1300        __u32 data_sz;
1301        void *data;
1302
1303        data = btf_get_raw_data(btf, &data_sz, btf->swapped_endian);
1304        if (!data)
1305                return errno = -ENOMEM, NULL;
1306
1307        btf->raw_size = data_sz;
1308        if (btf->swapped_endian)
1309                btf->raw_data_swapped = data;
1310        else
1311                btf->raw_data = data;
1312        *size = data_sz;
1313        return data;
1314}
1315
1316const char *btf__str_by_offset(const struct btf *btf, __u32 offset)
1317{
1318        if (offset < btf->start_str_off)
1319                return btf__str_by_offset(btf->base_btf, offset);
1320        else if (offset - btf->start_str_off < btf->hdr->str_len)
1321                return btf_strs_data(btf) + (offset - btf->start_str_off);
1322        else
1323                return errno = EINVAL, NULL;
1324}
1325
1326const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
1327{
1328        return btf__str_by_offset(btf, offset);
1329}
1330
1331struct btf *btf_get_from_fd(int btf_fd, struct btf *base_btf)
1332{
1333        struct bpf_btf_info btf_info;
1334        __u32 len = sizeof(btf_info);
1335        __u32 last_size;
1336        struct btf *btf;
1337        void *ptr;
1338        int err;
1339
1340        /* we won't know btf_size until we call bpf_obj_get_info_by_fd(). so
1341         * let's start with a sane default - 4KiB here - and resize it only if
1342         * bpf_obj_get_info_by_fd() needs a bigger buffer.
1343         */
1344        last_size = 4096;
1345        ptr = malloc(last_size);
1346        if (!ptr)
1347                return ERR_PTR(-ENOMEM);
1348
1349        memset(&btf_info, 0, sizeof(btf_info));
1350        btf_info.btf = ptr_to_u64(ptr);
1351        btf_info.btf_size = last_size;
1352        err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1353
1354        if (!err && btf_info.btf_size > last_size) {
1355                void *temp_ptr;
1356
1357                last_size = btf_info.btf_size;
1358                temp_ptr = realloc(ptr, last_size);
1359                if (!temp_ptr) {
1360                        btf = ERR_PTR(-ENOMEM);
1361                        goto exit_free;
1362                }
1363                ptr = temp_ptr;
1364
1365                len = sizeof(btf_info);
1366                memset(&btf_info, 0, sizeof(btf_info));
1367                btf_info.btf = ptr_to_u64(ptr);
1368                btf_info.btf_size = last_size;
1369
1370                err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1371        }
1372
1373        if (err || btf_info.btf_size > last_size) {
1374                btf = err ? ERR_PTR(-errno) : ERR_PTR(-E2BIG);
1375                goto exit_free;
1376        }
1377
1378        btf = btf_new(ptr, btf_info.btf_size, base_btf);
1379
1380exit_free:
1381        free(ptr);
1382        return btf;
1383}
1384
1385struct btf *btf__load_from_kernel_by_id_split(__u32 id, struct btf *base_btf)
1386{
1387        struct btf *btf;
1388        int btf_fd;
1389
1390        btf_fd = bpf_btf_get_fd_by_id(id);
1391        if (btf_fd < 0)
1392                return libbpf_err_ptr(-errno);
1393
1394        btf = btf_get_from_fd(btf_fd, base_btf);
1395        close(btf_fd);
1396
1397        return libbpf_ptr(btf);
1398}
1399
1400struct btf *btf__load_from_kernel_by_id(__u32 id)
1401{
1402        return btf__load_from_kernel_by_id_split(id, NULL);
1403}
1404
1405int btf__get_from_id(__u32 id, struct btf **btf)
1406{
1407        struct btf *res;
1408        int err;
1409
1410        *btf = NULL;
1411        res = btf__load_from_kernel_by_id(id);
1412        err = libbpf_get_error(res);
1413
1414        if (err)
1415                return libbpf_err(err);
1416
1417        *btf = res;
1418        return 0;
1419}
1420
1421int btf__get_map_kv_tids(const struct btf *btf, const char *map_name,
1422                         __u32 expected_key_size, __u32 expected_value_size,
1423                         __u32 *key_type_id, __u32 *value_type_id)
1424{
1425        const struct btf_type *container_type;
1426        const struct btf_member *key, *value;
1427        const size_t max_name = 256;
1428        char container_name[max_name];
1429        __s64 key_size, value_size;
1430        __s32 container_id;
1431
1432        if (snprintf(container_name, max_name, "____btf_map_%s", map_name) == max_name) {
1433                pr_warn("map:%s length of '____btf_map_%s' is too long\n",
1434                        map_name, map_name);
1435                return libbpf_err(-EINVAL);
1436        }
1437
1438        container_id = btf__find_by_name(btf, container_name);
1439        if (container_id < 0) {
1440                pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
1441                         map_name, container_name);
1442                return libbpf_err(container_id);
1443        }
1444
1445        container_type = btf__type_by_id(btf, container_id);
1446        if (!container_type) {
1447                pr_warn("map:%s cannot find BTF type for container_id:%u\n",
1448                        map_name, container_id);
1449                return libbpf_err(-EINVAL);
1450        }
1451
1452        if (!btf_is_struct(container_type) || btf_vlen(container_type) < 2) {
1453                pr_warn("map:%s container_name:%s is an invalid container struct\n",
1454                        map_name, container_name);
1455                return libbpf_err(-EINVAL);
1456        }
1457
1458        key = btf_members(container_type);
1459        value = key + 1;
1460
1461        key_size = btf__resolve_size(btf, key->type);
1462        if (key_size < 0) {
1463                pr_warn("map:%s invalid BTF key_type_size\n", map_name);
1464                return libbpf_err(key_size);
1465        }
1466
1467        if (expected_key_size != key_size) {
1468                pr_warn("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
1469                        map_name, (__u32)key_size, expected_key_size);
1470                return libbpf_err(-EINVAL);
1471        }
1472
1473        value_size = btf__resolve_size(btf, value->type);
1474        if (value_size < 0) {
1475                pr_warn("map:%s invalid BTF value_type_size\n", map_name);
1476                return libbpf_err(value_size);
1477        }
1478
1479        if (expected_value_size != value_size) {
1480                pr_warn("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
1481                        map_name, (__u32)value_size, expected_value_size);
1482                return libbpf_err(-EINVAL);
1483        }
1484
1485        *key_type_id = key->type;
1486        *value_type_id = value->type;
1487
1488        return 0;
1489}
1490
1491static void btf_invalidate_raw_data(struct btf *btf)
1492{
1493        if (btf->raw_data) {
1494                free(btf->raw_data);
1495                btf->raw_data = NULL;
1496        }
1497        if (btf->raw_data_swapped) {
1498                free(btf->raw_data_swapped);
1499                btf->raw_data_swapped = NULL;
1500        }
1501}
1502
1503/* Ensure BTF is ready to be modified (by splitting into a three memory
1504 * regions for header, types, and strings). Also invalidate cached
1505 * raw_data, if any.
1506 */
1507static int btf_ensure_modifiable(struct btf *btf)
1508{
1509        void *hdr, *types;
1510        struct strset *set = NULL;
1511        int err = -ENOMEM;
1512
1513        if (btf_is_modifiable(btf)) {
1514                /* any BTF modification invalidates raw_data */
1515                btf_invalidate_raw_data(btf);
1516                return 0;
1517        }
1518
1519        /* split raw data into three memory regions */
1520        hdr = malloc(btf->hdr->hdr_len);
1521        types = malloc(btf->hdr->type_len);
1522        if (!hdr || !types)
1523                goto err_out;
1524
1525        memcpy(hdr, btf->hdr, btf->hdr->hdr_len);
1526        memcpy(types, btf->types_data, btf->hdr->type_len);
1527
1528        /* build lookup index for all strings */
1529        set = strset__new(BTF_MAX_STR_OFFSET, btf->strs_data, btf->hdr->str_len);
1530        if (IS_ERR(set)) {
1531                err = PTR_ERR(set);
1532                goto err_out;
1533        }
1534
1535        /* only when everything was successful, update internal state */
1536        btf->hdr = hdr;
1537        btf->types_data = types;
1538        btf->types_data_cap = btf->hdr->type_len;
1539        btf->strs_data = NULL;
1540        btf->strs_set = set;
1541        /* if BTF was created from scratch, all strings are guaranteed to be
1542         * unique and deduplicated
1543         */
1544        if (btf->hdr->str_len == 0)
1545                btf->strs_deduped = true;
1546        if (!btf->base_btf && btf->hdr->str_len == 1)
1547                btf->strs_deduped = true;
1548
1549        /* invalidate raw_data representation */
1550        btf_invalidate_raw_data(btf);
1551
1552        return 0;
1553
1554err_out:
1555        strset__free(set);
1556        free(hdr);
1557        free(types);
1558        return err;
1559}
1560
1561/* Find an offset in BTF string section that corresponds to a given string *s*.
1562 * Returns:
1563 *   - >0 offset into string section, if string is found;
1564 *   - -ENOENT, if string is not in the string section;
1565 *   - <0, on any other error.
1566 */
1567int btf__find_str(struct btf *btf, const char *s)
1568{
1569        int off;
1570
1571        if (btf->base_btf) {
1572                off = btf__find_str(btf->base_btf, s);
1573                if (off != -ENOENT)
1574                        return off;
1575        }
1576
1577        /* BTF needs to be in a modifiable state to build string lookup index */
1578        if (btf_ensure_modifiable(btf))
1579                return libbpf_err(-ENOMEM);
1580
1581        off = strset__find_str(btf->strs_set, s);
1582        if (off < 0)
1583                return libbpf_err(off);
1584
1585        return btf->start_str_off + off;
1586}
1587
1588/* Add a string s to the BTF string section.
1589 * Returns:
1590 *   - > 0 offset into string section, on success;
1591 *   - < 0, on error.
1592 */
1593int btf__add_str(struct btf *btf, const char *s)
1594{
1595        int off;
1596
1597        if (btf->base_btf) {
1598                off = btf__find_str(btf->base_btf, s);
1599                if (off != -ENOENT)
1600                        return off;
1601        }
1602
1603        if (btf_ensure_modifiable(btf))
1604                return libbpf_err(-ENOMEM);
1605
1606        off = strset__add_str(btf->strs_set, s);
1607        if (off < 0)
1608                return libbpf_err(off);
1609
1610        btf->hdr->str_len = strset__data_size(btf->strs_set);
1611
1612        return btf->start_str_off + off;
1613}
1614
1615static void *btf_add_type_mem(struct btf *btf, size_t add_sz)
1616{
1617        return libbpf_add_mem(&btf->types_data, &btf->types_data_cap, 1,
1618                              btf->hdr->type_len, UINT_MAX, add_sz);
1619}
1620
1621static void btf_type_inc_vlen(struct btf_type *t)
1622{
1623        t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, btf_kflag(t));
1624}
1625
1626static int btf_commit_type(struct btf *btf, int data_sz)
1627{
1628        int err;
1629
1630        err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1631        if (err)
1632                return libbpf_err(err);
1633
1634        btf->hdr->type_len += data_sz;
1635        btf->hdr->str_off += data_sz;
1636        btf->nr_types++;
1637        return btf->start_id + btf->nr_types - 1;
1638}
1639
1640struct btf_pipe {
1641        const struct btf *src;
1642        struct btf *dst;
1643};
1644
1645static int btf_rewrite_str(__u32 *str_off, void *ctx)
1646{
1647        struct btf_pipe *p = ctx;
1648        int off;
1649
1650        if (!*str_off) /* nothing to do for empty strings */
1651                return 0;
1652
1653        off = btf__add_str(p->dst, btf__str_by_offset(p->src, *str_off));
1654        if (off < 0)
1655                return off;
1656
1657        *str_off = off;
1658        return 0;
1659}
1660
1661int btf__add_type(struct btf *btf, const struct btf *src_btf, const struct btf_type *src_type)
1662{
1663        struct btf_pipe p = { .src = src_btf, .dst = btf };
1664        struct btf_type *t;
1665        int sz, err;
1666
1667        sz = btf_type_size(src_type);
1668        if (sz < 0)
1669                return libbpf_err(sz);
1670
1671        /* deconstruct BTF, if necessary, and invalidate raw_data */
1672        if (btf_ensure_modifiable(btf))
1673                return libbpf_err(-ENOMEM);
1674
1675        t = btf_add_type_mem(btf, sz);
1676        if (!t)
1677                return libbpf_err(-ENOMEM);
1678
1679        memcpy(t, src_type, sz);
1680
1681        err = btf_type_visit_str_offs(t, btf_rewrite_str, &p);
1682        if (err)
1683                return libbpf_err(err);
1684
1685        return btf_commit_type(btf, sz);
1686}
1687
1688/*
1689 * Append new BTF_KIND_INT type with:
1690 *   - *name* - non-empty, non-NULL type name;
1691 *   - *sz* - power-of-2 (1, 2, 4, ..) size of the type, in bytes;
1692 *   - encoding is a combination of BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL.
1693 * Returns:
1694 *   - >0, type ID of newly added BTF type;
1695 *   - <0, on error.
1696 */
1697int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding)
1698{
1699        struct btf_type *t;
1700        int sz, name_off;
1701
1702        /* non-empty name */
1703        if (!name || !name[0])
1704                return libbpf_err(-EINVAL);
1705        /* byte_sz must be power of 2 */
1706        if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
1707                return libbpf_err(-EINVAL);
1708        if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
1709                return libbpf_err(-EINVAL);
1710
1711        /* deconstruct BTF, if necessary, and invalidate raw_data */
1712        if (btf_ensure_modifiable(btf))
1713                return libbpf_err(-ENOMEM);
1714
1715        sz = sizeof(struct btf_type) + sizeof(int);
1716        t = btf_add_type_mem(btf, sz);
1717        if (!t)
1718                return libbpf_err(-ENOMEM);
1719
1720        /* if something goes wrong later, we might end up with an extra string,
1721         * but that shouldn't be a problem, because BTF can't be constructed
1722         * completely anyway and will most probably be just discarded
1723         */
1724        name_off = btf__add_str(btf, name);
1725        if (name_off < 0)
1726                return name_off;
1727
1728        t->name_off = name_off;
1729        t->info = btf_type_info(BTF_KIND_INT, 0, 0);
1730        t->size = byte_sz;
1731        /* set INT info, we don't allow setting legacy bit offset/size */
1732        *(__u32 *)(t + 1) = (encoding << 24) | (byte_sz * 8);
1733
1734        return btf_commit_type(btf, sz);
1735}
1736
1737/*
1738 * Append new BTF_KIND_FLOAT type with:
1739 *   - *name* - non-empty, non-NULL type name;
1740 *   - *sz* - size of the type, in bytes;
1741 * Returns:
1742 *   - >0, type ID of newly added BTF type;
1743 *   - <0, on error.
1744 */
1745int btf__add_float(struct btf *btf, const char *name, size_t byte_sz)
1746{
1747        struct btf_type *t;
1748        int sz, name_off;
1749
1750        /* non-empty name */
1751        if (!name || !name[0])
1752                return libbpf_err(-EINVAL);
1753
1754        /* byte_sz must be one of the explicitly allowed values */
1755        if (byte_sz != 2 && byte_sz != 4 && byte_sz != 8 && byte_sz != 12 &&
1756            byte_sz != 16)
1757                return libbpf_err(-EINVAL);
1758
1759        if (btf_ensure_modifiable(btf))
1760                return libbpf_err(-ENOMEM);
1761
1762        sz = sizeof(struct btf_type);
1763        t = btf_add_type_mem(btf, sz);
1764        if (!t)
1765                return libbpf_err(-ENOMEM);
1766
1767        name_off = btf__add_str(btf, name);
1768        if (name_off < 0)
1769                return name_off;
1770
1771        t->name_off = name_off;
1772        t->info = btf_type_info(BTF_KIND_FLOAT, 0, 0);
1773        t->size = byte_sz;
1774
1775        return btf_commit_type(btf, sz);
1776}
1777
1778/* it's completely legal to append BTF types with type IDs pointing forward to
1779 * types that haven't been appended yet, so we only make sure that id looks
1780 * sane, we can't guarantee that ID will always be valid
1781 */
1782static int validate_type_id(int id)
1783{
1784        if (id < 0 || id > BTF_MAX_NR_TYPES)
1785                return -EINVAL;
1786        return 0;
1787}
1788
1789/* generic append function for PTR, TYPEDEF, CONST/VOLATILE/RESTRICT */
1790static int btf_add_ref_kind(struct btf *btf, int kind, const char *name, int ref_type_id)
1791{
1792        struct btf_type *t;
1793        int sz, name_off = 0;
1794
1795        if (validate_type_id(ref_type_id))
1796                return libbpf_err(-EINVAL);
1797
1798        if (btf_ensure_modifiable(btf))
1799                return libbpf_err(-ENOMEM);
1800
1801        sz = sizeof(struct btf_type);
1802        t = btf_add_type_mem(btf, sz);
1803        if (!t)
1804                return libbpf_err(-ENOMEM);
1805
1806        if (name && name[0]) {
1807                name_off = btf__add_str(btf, name);
1808                if (name_off < 0)
1809                        return name_off;
1810        }
1811
1812        t->name_off = name_off;
1813        t->info = btf_type_info(kind, 0, 0);
1814        t->type = ref_type_id;
1815
1816        return btf_commit_type(btf, sz);
1817}
1818
1819/*
1820 * Append new BTF_KIND_PTR type with:
1821 *   - *ref_type_id* - referenced type ID, it might not exist yet;
1822 * Returns:
1823 *   - >0, type ID of newly added BTF type;
1824 *   - <0, on error.
1825 */
1826int btf__add_ptr(struct btf *btf, int ref_type_id)
1827{
1828        return btf_add_ref_kind(btf, BTF_KIND_PTR, NULL, ref_type_id);
1829}
1830
1831/*
1832 * Append new BTF_KIND_ARRAY type with:
1833 *   - *index_type_id* - type ID of the type describing array index;
1834 *   - *elem_type_id* - type ID of the type describing array element;
1835 *   - *nr_elems* - the size of the array;
1836 * Returns:
1837 *   - >0, type ID of newly added BTF type;
1838 *   - <0, on error.
1839 */
1840int btf__add_array(struct btf *btf, int index_type_id, int elem_type_id, __u32 nr_elems)
1841{
1842        struct btf_type *t;
1843        struct btf_array *a;
1844        int sz;
1845
1846        if (validate_type_id(index_type_id) || validate_type_id(elem_type_id))
1847                return libbpf_err(-EINVAL);
1848
1849        if (btf_ensure_modifiable(btf))
1850                return libbpf_err(-ENOMEM);
1851
1852        sz = sizeof(struct btf_type) + sizeof(struct btf_array);
1853        t = btf_add_type_mem(btf, sz);
1854        if (!t)
1855                return libbpf_err(-ENOMEM);
1856
1857        t->name_off = 0;
1858        t->info = btf_type_info(BTF_KIND_ARRAY, 0, 0);
1859        t->size = 0;
1860
1861        a = btf_array(t);
1862        a->type = elem_type_id;
1863        a->index_type = index_type_id;
1864        a->nelems = nr_elems;
1865
1866        return btf_commit_type(btf, sz);
1867}
1868
1869/* generic STRUCT/UNION append function */
1870static int btf_add_composite(struct btf *btf, int kind, const char *name, __u32 bytes_sz)
1871{
1872        struct btf_type *t;
1873        int sz, name_off = 0;
1874
1875        if (btf_ensure_modifiable(btf))
1876                return libbpf_err(-ENOMEM);
1877
1878        sz = sizeof(struct btf_type);
1879        t = btf_add_type_mem(btf, sz);
1880        if (!t)
1881                return libbpf_err(-ENOMEM);
1882
1883        if (name && name[0]) {
1884                name_off = btf__add_str(btf, name);
1885                if (name_off < 0)
1886                        return name_off;
1887        }
1888
1889        /* start out with vlen=0 and no kflag; this will be adjusted when
1890         * adding each member
1891         */
1892        t->name_off = name_off;
1893        t->info = btf_type_info(kind, 0, 0);
1894        t->size = bytes_sz;
1895
1896        return btf_commit_type(btf, sz);
1897}
1898
1899/*
1900 * Append new BTF_KIND_STRUCT type with:
1901 *   - *name* - name of the struct, can be NULL or empty for anonymous structs;
1902 *   - *byte_sz* - size of the struct, in bytes;
1903 *
1904 * Struct initially has no fields in it. Fields can be added by
1905 * btf__add_field() right after btf__add_struct() succeeds.
1906 *
1907 * Returns:
1908 *   - >0, type ID of newly added BTF type;
1909 *   - <0, on error.
1910 */
1911int btf__add_struct(struct btf *btf, const char *name, __u32 byte_sz)
1912{
1913        return btf_add_composite(btf, BTF_KIND_STRUCT, name, byte_sz);
1914}
1915
1916/*
1917 * Append new BTF_KIND_UNION type with:
1918 *   - *name* - name of the union, can be NULL or empty for anonymous union;
1919 *   - *byte_sz* - size of the union, in bytes;
1920 *
1921 * Union initially has no fields in it. Fields can be added by
1922 * btf__add_field() right after btf__add_union() succeeds. All fields
1923 * should have *bit_offset* of 0.
1924 *
1925 * Returns:
1926 *   - >0, type ID of newly added BTF type;
1927 *   - <0, on error.
1928 */
1929int btf__add_union(struct btf *btf, const char *name, __u32 byte_sz)
1930{
1931        return btf_add_composite(btf, BTF_KIND_UNION, name, byte_sz);
1932}
1933
1934static struct btf_type *btf_last_type(struct btf *btf)
1935{
1936        return btf_type_by_id(btf, btf__get_nr_types(btf));
1937}
1938
1939/*
1940 * Append new field for the current STRUCT/UNION type with:
1941 *   - *name* - name of the field, can be NULL or empty for anonymous field;
1942 *   - *type_id* - type ID for the type describing field type;
1943 *   - *bit_offset* - bit offset of the start of the field within struct/union;
1944 *   - *bit_size* - bit size of a bitfield, 0 for non-bitfield fields;
1945 * Returns:
1946 *   -  0, on success;
1947 *   - <0, on error.
1948 */
1949int btf__add_field(struct btf *btf, const char *name, int type_id,
1950                   __u32 bit_offset, __u32 bit_size)
1951{
1952        struct btf_type *t;
1953        struct btf_member *m;
1954        bool is_bitfield;
1955        int sz, name_off = 0;
1956
1957        /* last type should be union/struct */
1958        if (btf->nr_types == 0)
1959                return libbpf_err(-EINVAL);
1960        t = btf_last_type(btf);
1961        if (!btf_is_composite(t))
1962                return libbpf_err(-EINVAL);
1963
1964        if (validate_type_id(type_id))
1965                return libbpf_err(-EINVAL);
1966        /* best-effort bit field offset/size enforcement */
1967        is_bitfield = bit_size || (bit_offset % 8 != 0);
1968        if (is_bitfield && (bit_size == 0 || bit_size > 255 || bit_offset > 0xffffff))
1969                return libbpf_err(-EINVAL);
1970
1971        /* only offset 0 is allowed for unions */
1972        if (btf_is_union(t) && bit_offset)
1973                return libbpf_err(-EINVAL);
1974
1975        /* decompose and invalidate raw data */
1976        if (btf_ensure_modifiable(btf))
1977                return libbpf_err(-ENOMEM);
1978
1979        sz = sizeof(struct btf_member);
1980        m = btf_add_type_mem(btf, sz);
1981        if (!m)
1982                return libbpf_err(-ENOMEM);
1983
1984        if (name && name[0]) {
1985                name_off = btf__add_str(btf, name);
1986                if (name_off < 0)
1987                        return name_off;
1988        }
1989
1990        m->name_off = name_off;
1991        m->type = type_id;
1992        m->offset = bit_offset | (bit_size << 24);
1993
1994        /* btf_add_type_mem can invalidate t pointer */
1995        t = btf_last_type(btf);
1996        /* update parent type's vlen and kflag */
1997        t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, is_bitfield || btf_kflag(t));
1998
1999        btf->hdr->type_len += sz;
2000        btf->hdr->str_off += sz;
2001        return 0;
2002}
2003
2004/*
2005 * Append new BTF_KIND_ENUM type with:
2006 *   - *name* - name of the enum, can be NULL or empty for anonymous enums;
2007 *   - *byte_sz* - size of the enum, in bytes.
2008 *
2009 * Enum initially has no enum values in it (and corresponds to enum forward
2010 * declaration). Enumerator values can be added by btf__add_enum_value()
2011 * immediately after btf__add_enum() succeeds.
2012 *
2013 * Returns:
2014 *   - >0, type ID of newly added BTF type;
2015 *   - <0, on error.
2016 */
2017int btf__add_enum(struct btf *btf, const char *name, __u32 byte_sz)
2018{
2019        struct btf_type *t;
2020        int sz, name_off = 0;
2021
2022        /* byte_sz must be power of 2 */
2023        if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 8)
2024                return libbpf_err(-EINVAL);
2025
2026        if (btf_ensure_modifiable(btf))
2027                return libbpf_err(-ENOMEM);
2028
2029        sz = sizeof(struct btf_type);
2030        t = btf_add_type_mem(btf, sz);
2031        if (!t)
2032                return libbpf_err(-ENOMEM);
2033
2034        if (name && name[0]) {
2035                name_off = btf__add_str(btf, name);
2036                if (name_off < 0)
2037                        return name_off;
2038        }
2039
2040        /* start out with vlen=0; it will be adjusted when adding enum values */
2041        t->name_off = name_off;
2042        t->info = btf_type_info(BTF_KIND_ENUM, 0, 0);
2043        t->size = byte_sz;
2044
2045        return btf_commit_type(btf, sz);
2046}
2047
2048/*
2049 * Append new enum value for the current ENUM type with:
2050 *   - *name* - name of the enumerator value, can't be NULL or empty;
2051 *   - *value* - integer value corresponding to enum value *name*;
2052 * Returns:
2053 *   -  0, on success;
2054 *   - <0, on error.
2055 */
2056int btf__add_enum_value(struct btf *btf, const char *name, __s64 value)
2057{
2058        struct btf_type *t;
2059        struct btf_enum *v;
2060        int sz, name_off;
2061
2062        /* last type should be BTF_KIND_ENUM */
2063        if (btf->nr_types == 0)
2064                return libbpf_err(-EINVAL);
2065        t = btf_last_type(btf);
2066        if (!btf_is_enum(t))
2067                return libbpf_err(-EINVAL);
2068
2069        /* non-empty name */
2070        if (!name || !name[0])
2071                return libbpf_err(-EINVAL);
2072        if (value < INT_MIN || value > UINT_MAX)
2073                return libbpf_err(-E2BIG);
2074
2075        /* decompose and invalidate raw data */
2076        if (btf_ensure_modifiable(btf))
2077                return libbpf_err(-ENOMEM);
2078
2079        sz = sizeof(struct btf_enum);
2080        v = btf_add_type_mem(btf, sz);
2081        if (!v)
2082                return libbpf_err(-ENOMEM);
2083
2084        name_off = btf__add_str(btf, name);
2085        if (name_off < 0)
2086                return name_off;
2087
2088        v->name_off = name_off;
2089        v->val = value;
2090
2091        /* update parent type's vlen */
2092        t = btf_last_type(btf);
2093        btf_type_inc_vlen(t);
2094
2095        btf->hdr->type_len += sz;
2096        btf->hdr->str_off += sz;
2097        return 0;
2098}
2099
2100/*
2101 * Append new BTF_KIND_FWD type with:
2102 *   - *name*, non-empty/non-NULL name;
2103 *   - *fwd_kind*, kind of forward declaration, one of BTF_FWD_STRUCT,
2104 *     BTF_FWD_UNION, or BTF_FWD_ENUM;
2105 * Returns:
2106 *   - >0, type ID of newly added BTF type;
2107 *   - <0, on error.
2108 */
2109int btf__add_fwd(struct btf *btf, const char *name, enum btf_fwd_kind fwd_kind)
2110{
2111        if (!name || !name[0])
2112                return libbpf_err(-EINVAL);
2113
2114        switch (fwd_kind) {
2115        case BTF_FWD_STRUCT:
2116        case BTF_FWD_UNION: {
2117                struct btf_type *t;
2118                int id;
2119
2120                id = btf_add_ref_kind(btf, BTF_KIND_FWD, name, 0);
2121                if (id <= 0)
2122                        return id;
2123                t = btf_type_by_id(btf, id);
2124                t->info = btf_type_info(BTF_KIND_FWD, 0, fwd_kind == BTF_FWD_UNION);
2125                return id;
2126        }
2127        case BTF_FWD_ENUM:
2128                /* enum forward in BTF currently is just an enum with no enum
2129                 * values; we also assume a standard 4-byte size for it
2130                 */
2131                return btf__add_enum(btf, name, sizeof(int));
2132        default:
2133                return libbpf_err(-EINVAL);
2134        }
2135}
2136
2137/*
2138 * Append new BTF_KING_TYPEDEF type with:
2139 *   - *name*, non-empty/non-NULL name;
2140 *   - *ref_type_id* - referenced type ID, it might not exist yet;
2141 * Returns:
2142 *   - >0, type ID of newly added BTF type;
2143 *   - <0, on error.
2144 */
2145int btf__add_typedef(struct btf *btf, const char *name, int ref_type_id)
2146{
2147        if (!name || !name[0])
2148                return libbpf_err(-EINVAL);
2149
2150        return btf_add_ref_kind(btf, BTF_KIND_TYPEDEF, name, ref_type_id);
2151}
2152
2153/*
2154 * Append new BTF_KIND_VOLATILE type with:
2155 *   - *ref_type_id* - referenced type ID, it might not exist yet;
2156 * Returns:
2157 *   - >0, type ID of newly added BTF type;
2158 *   - <0, on error.
2159 */
2160int btf__add_volatile(struct btf *btf, int ref_type_id)
2161{
2162        return btf_add_ref_kind(btf, BTF_KIND_VOLATILE, NULL, ref_type_id);
2163}
2164
2165/*
2166 * Append new BTF_KIND_CONST type with:
2167 *   - *ref_type_id* - referenced type ID, it might not exist yet;
2168 * Returns:
2169 *   - >0, type ID of newly added BTF type;
2170 *   - <0, on error.
2171 */
2172int btf__add_const(struct btf *btf, int ref_type_id)
2173{
2174        return btf_add_ref_kind(btf, BTF_KIND_CONST, NULL, ref_type_id);
2175}
2176
2177/*
2178 * Append new BTF_KIND_RESTRICT type with:
2179 *   - *ref_type_id* - referenced type ID, it might not exist yet;
2180 * Returns:
2181 *   - >0, type ID of newly added BTF type;
2182 *   - <0, on error.
2183 */
2184int btf__add_restrict(struct btf *btf, int ref_type_id)
2185{
2186        return btf_add_ref_kind(btf, BTF_KIND_RESTRICT, NULL, ref_type_id);
2187}
2188
2189/*
2190 * Append new BTF_KIND_FUNC type with:
2191 *   - *name*, non-empty/non-NULL name;
2192 *   - *proto_type_id* - FUNC_PROTO's type ID, it might not exist yet;
2193 * Returns:
2194 *   - >0, type ID of newly added BTF type;
2195 *   - <0, on error.
2196 */
2197int btf__add_func(struct btf *btf, const char *name,
2198                  enum btf_func_linkage linkage, int proto_type_id)
2199{
2200        int id;
2201
2202        if (!name || !name[0])
2203                return libbpf_err(-EINVAL);
2204        if (linkage != BTF_FUNC_STATIC && linkage != BTF_FUNC_GLOBAL &&
2205            linkage != BTF_FUNC_EXTERN)
2206                return libbpf_err(-EINVAL);
2207
2208        id = btf_add_ref_kind(btf, BTF_KIND_FUNC, name, proto_type_id);
2209        if (id > 0) {
2210                struct btf_type *t = btf_type_by_id(btf, id);
2211
2212                t->info = btf_type_info(BTF_KIND_FUNC, linkage, 0);
2213        }
2214        return libbpf_err(id);
2215}
2216
2217/*
2218 * Append new BTF_KIND_FUNC_PROTO with:
2219 *   - *ret_type_id* - type ID for return result of a function.
2220 *
2221 * Function prototype initially has no arguments, but they can be added by
2222 * btf__add_func_param() one by one, immediately after
2223 * btf__add_func_proto() succeeded.
2224 *
2225 * Returns:
2226 *   - >0, type ID of newly added BTF type;
2227 *   - <0, on error.
2228 */
2229int btf__add_func_proto(struct btf *btf, int ret_type_id)
2230{
2231        struct btf_type *t;
2232        int sz;
2233
2234        if (validate_type_id(ret_type_id))
2235                return libbpf_err(-EINVAL);
2236
2237        if (btf_ensure_modifiable(btf))
2238                return libbpf_err(-ENOMEM);
2239
2240        sz = sizeof(struct btf_type);
2241        t = btf_add_type_mem(btf, sz);
2242        if (!t)
2243                return libbpf_err(-ENOMEM);
2244
2245        /* start out with vlen=0; this will be adjusted when adding enum
2246         * values, if necessary
2247         */
2248        t->name_off = 0;
2249        t->info = btf_type_info(BTF_KIND_FUNC_PROTO, 0, 0);
2250        t->type = ret_type_id;
2251
2252        return btf_commit_type(btf, sz);
2253}
2254
2255/*
2256 * Append new function parameter for current FUNC_PROTO type with:
2257 *   - *name* - parameter name, can be NULL or empty;
2258 *   - *type_id* - type ID describing the type of the parameter.
2259 * Returns:
2260 *   -  0, on success;
2261 *   - <0, on error.
2262 */
2263int btf__add_func_param(struct btf *btf, const char *name, int type_id)
2264{
2265        struct btf_type *t;
2266        struct btf_param *p;
2267        int sz, name_off = 0;
2268
2269        if (validate_type_id(type_id))
2270                return libbpf_err(-EINVAL);
2271
2272        /* last type should be BTF_KIND_FUNC_PROTO */
2273        if (btf->nr_types == 0)
2274                return libbpf_err(-EINVAL);
2275        t = btf_last_type(btf);
2276        if (!btf_is_func_proto(t))
2277                return libbpf_err(-EINVAL);
2278
2279        /* decompose and invalidate raw data */
2280        if (btf_ensure_modifiable(btf))
2281                return libbpf_err(-ENOMEM);
2282
2283        sz = sizeof(struct btf_param);
2284        p = btf_add_type_mem(btf, sz);
2285        if (!p)
2286                return libbpf_err(-ENOMEM);
2287
2288        if (name && name[0]) {
2289                name_off = btf__add_str(btf, name);
2290                if (name_off < 0)
2291                        return name_off;
2292        }
2293
2294        p->name_off = name_off;
2295        p->type = type_id;
2296
2297        /* update parent type's vlen */
2298        t = btf_last_type(btf);
2299        btf_type_inc_vlen(t);
2300
2301        btf->hdr->type_len += sz;
2302        btf->hdr->str_off += sz;
2303        return 0;
2304}
2305
2306/*
2307 * Append new BTF_KIND_VAR type with:
2308 *   - *name* - non-empty/non-NULL name;
2309 *   - *linkage* - variable linkage, one of BTF_VAR_STATIC,
2310 *     BTF_VAR_GLOBAL_ALLOCATED, or BTF_VAR_GLOBAL_EXTERN;
2311 *   - *type_id* - type ID of the type describing the type of the variable.
2312 * Returns:
2313 *   - >0, type ID of newly added BTF type;
2314 *   - <0, on error.
2315 */
2316int btf__add_var(struct btf *btf, const char *name, int linkage, int type_id)
2317{
2318        struct btf_type *t;
2319        struct btf_var *v;
2320        int sz, name_off;
2321
2322        /* non-empty name */
2323        if (!name || !name[0])
2324                return libbpf_err(-EINVAL);
2325        if (linkage != BTF_VAR_STATIC && linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2326            linkage != BTF_VAR_GLOBAL_EXTERN)
2327                return libbpf_err(-EINVAL);
2328        if (validate_type_id(type_id))
2329                return libbpf_err(-EINVAL);
2330
2331        /* deconstruct BTF, if necessary, and invalidate raw_data */
2332        if (btf_ensure_modifiable(btf))
2333                return libbpf_err(-ENOMEM);
2334
2335        sz = sizeof(struct btf_type) + sizeof(struct btf_var);
2336        t = btf_add_type_mem(btf, sz);
2337        if (!t)
2338                return libbpf_err(-ENOMEM);
2339
2340        name_off = btf__add_str(btf, name);
2341        if (name_off < 0)
2342                return name_off;
2343
2344        t->name_off = name_off;
2345        t->info = btf_type_info(BTF_KIND_VAR, 0, 0);
2346        t->type = type_id;
2347
2348        v = btf_var(t);
2349        v->linkage = linkage;
2350
2351        return btf_commit_type(btf, sz);
2352}
2353
2354/*
2355 * Append new BTF_KIND_DATASEC type with:
2356 *   - *name* - non-empty/non-NULL name;
2357 *   - *byte_sz* - data section size, in bytes.
2358 *
2359 * Data section is initially empty. Variables info can be added with
2360 * btf__add_datasec_var_info() calls, after btf__add_datasec() succeeds.
2361 *
2362 * Returns:
2363 *   - >0, type ID of newly added BTF type;
2364 *   - <0, on error.
2365 */
2366int btf__add_datasec(struct btf *btf, const char *name, __u32 byte_sz)
2367{
2368        struct btf_type *t;
2369        int sz, name_off;
2370
2371        /* non-empty name */
2372        if (!name || !name[0])
2373                return libbpf_err(-EINVAL);
2374
2375        if (btf_ensure_modifiable(btf))
2376                return libbpf_err(-ENOMEM);
2377
2378        sz = sizeof(struct btf_type);
2379        t = btf_add_type_mem(btf, sz);
2380        if (!t)
2381                return libbpf_err(-ENOMEM);
2382
2383        name_off = btf__add_str(btf, name);
2384        if (name_off < 0)
2385                return name_off;
2386
2387        /* start with vlen=0, which will be update as var_secinfos are added */
2388        t->name_off = name_off;
2389        t->info = btf_type_info(BTF_KIND_DATASEC, 0, 0);
2390        t->size = byte_sz;
2391
2392        return btf_commit_type(btf, sz);
2393}
2394
2395/*
2396 * Append new data section variable information entry for current DATASEC type:
2397 *   - *var_type_id* - type ID, describing type of the variable;
2398 *   - *offset* - variable offset within data section, in bytes;
2399 *   - *byte_sz* - variable size, in bytes.
2400 *
2401 * Returns:
2402 *   -  0, on success;
2403 *   - <0, on error.
2404 */
2405int btf__add_datasec_var_info(struct btf *btf, int var_type_id, __u32 offset, __u32 byte_sz)
2406{
2407        struct btf_type *t;
2408        struct btf_var_secinfo *v;
2409        int sz;
2410
2411        /* last type should be BTF_KIND_DATASEC */
2412        if (btf->nr_types == 0)
2413                return libbpf_err(-EINVAL);
2414        t = btf_last_type(btf);
2415        if (!btf_is_datasec(t))
2416                return libbpf_err(-EINVAL);
2417
2418        if (validate_type_id(var_type_id))
2419                return libbpf_err(-EINVAL);
2420
2421        /* decompose and invalidate raw data */
2422        if (btf_ensure_modifiable(btf))
2423                return libbpf_err(-ENOMEM);
2424
2425        sz = sizeof(struct btf_var_secinfo);
2426        v = btf_add_type_mem(btf, sz);
2427        if (!v)
2428                return libbpf_err(-ENOMEM);
2429
2430        v->type = var_type_id;
2431        v->offset = offset;
2432        v->size = byte_sz;
2433
2434        /* update parent type's vlen */
2435        t = btf_last_type(btf);
2436        btf_type_inc_vlen(t);
2437
2438        btf->hdr->type_len += sz;
2439        btf->hdr->str_off += sz;
2440        return 0;
2441}
2442
2443struct btf_ext_sec_setup_param {
2444        __u32 off;
2445        __u32 len;
2446        __u32 min_rec_size;
2447        struct btf_ext_info *ext_info;
2448        const char *desc;
2449};
2450
2451static int btf_ext_setup_info(struct btf_ext *btf_ext,
2452                              struct btf_ext_sec_setup_param *ext_sec)
2453{
2454        const struct btf_ext_info_sec *sinfo;
2455        struct btf_ext_info *ext_info;
2456        __u32 info_left, record_size;
2457        /* The start of the info sec (including the __u32 record_size). */
2458        void *info;
2459
2460        if (ext_sec->len == 0)
2461                return 0;
2462
2463        if (ext_sec->off & 0x03) {
2464                pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
2465                     ext_sec->desc);
2466                return -EINVAL;
2467        }
2468
2469        info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
2470        info_left = ext_sec->len;
2471
2472        if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
2473                pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
2474                         ext_sec->desc, ext_sec->off, ext_sec->len);
2475                return -EINVAL;
2476        }
2477
2478        /* At least a record size */
2479        if (info_left < sizeof(__u32)) {
2480                pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
2481                return -EINVAL;
2482        }
2483
2484        /* The record size needs to meet the minimum standard */
2485        record_size = *(__u32 *)info;
2486        if (record_size < ext_sec->min_rec_size ||
2487            record_size & 0x03) {
2488                pr_debug("%s section in .BTF.ext has invalid record size %u\n",
2489                         ext_sec->desc, record_size);
2490                return -EINVAL;
2491        }
2492
2493        sinfo = info + sizeof(__u32);
2494        info_left -= sizeof(__u32);
2495
2496        /* If no records, return failure now so .BTF.ext won't be used. */
2497        if (!info_left) {
2498                pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
2499                return -EINVAL;
2500        }
2501
2502        while (info_left) {
2503                unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
2504                __u64 total_record_size;
2505                __u32 num_records;
2506
2507                if (info_left < sec_hdrlen) {
2508                        pr_debug("%s section header is not found in .BTF.ext\n",
2509                             ext_sec->desc);
2510                        return -EINVAL;
2511                }
2512
2513                num_records = sinfo->num_info;
2514                if (num_records == 0) {
2515                        pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2516                             ext_sec->desc);
2517                        return -EINVAL;
2518                }
2519
2520                total_record_size = sec_hdrlen +
2521                                    (__u64)num_records * record_size;
2522                if (info_left < total_record_size) {
2523                        pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2524                             ext_sec->desc);
2525                        return -EINVAL;
2526                }
2527
2528                info_left -= total_record_size;
2529                sinfo = (void *)sinfo + total_record_size;
2530        }
2531
2532        ext_info = ext_sec->ext_info;
2533        ext_info->len = ext_sec->len - sizeof(__u32);
2534        ext_info->rec_size = record_size;
2535        ext_info->info = info + sizeof(__u32);
2536
2537        return 0;
2538}
2539
2540static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
2541{
2542        struct btf_ext_sec_setup_param param = {
2543                .off = btf_ext->hdr->func_info_off,
2544                .len = btf_ext->hdr->func_info_len,
2545                .min_rec_size = sizeof(struct bpf_func_info_min),
2546                .ext_info = &btf_ext->func_info,
2547                .desc = "func_info"
2548        };
2549
2550        return btf_ext_setup_info(btf_ext, &param);
2551}
2552
2553static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
2554{
2555        struct btf_ext_sec_setup_param param = {
2556                .off = btf_ext->hdr->line_info_off,
2557                .len = btf_ext->hdr->line_info_len,
2558                .min_rec_size = sizeof(struct bpf_line_info_min),
2559                .ext_info = &btf_ext->line_info,
2560                .desc = "line_info",
2561        };
2562
2563        return btf_ext_setup_info(btf_ext, &param);
2564}
2565
2566static int btf_ext_setup_core_relos(struct btf_ext *btf_ext)
2567{
2568        struct btf_ext_sec_setup_param param = {
2569                .off = btf_ext->hdr->core_relo_off,
2570                .len = btf_ext->hdr->core_relo_len,
2571                .min_rec_size = sizeof(struct bpf_core_relo),
2572                .ext_info = &btf_ext->core_relo_info,
2573                .desc = "core_relo",
2574        };
2575
2576        return btf_ext_setup_info(btf_ext, &param);
2577}
2578
2579static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
2580{
2581        const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
2582
2583        if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
2584            data_size < hdr->hdr_len) {
2585                pr_debug("BTF.ext header not found");
2586                return -EINVAL;
2587        }
2588
2589        if (hdr->magic == bswap_16(BTF_MAGIC)) {
2590                pr_warn("BTF.ext in non-native endianness is not supported\n");
2591                return -ENOTSUP;
2592        } else if (hdr->magic != BTF_MAGIC) {
2593                pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
2594                return -EINVAL;
2595        }
2596
2597        if (hdr->version != BTF_VERSION) {
2598                pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
2599                return -ENOTSUP;
2600        }
2601
2602        if (hdr->flags) {
2603                pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
2604                return -ENOTSUP;
2605        }
2606
2607        if (data_size == hdr->hdr_len) {
2608                pr_debug("BTF.ext has no data\n");
2609                return -EINVAL;
2610        }
2611
2612        return 0;
2613}
2614
2615void btf_ext__free(struct btf_ext *btf_ext)
2616{
2617        if (IS_ERR_OR_NULL(btf_ext))
2618                return;
2619        free(btf_ext->data);
2620        free(btf_ext);
2621}
2622
2623struct btf_ext *btf_ext__new(__u8 *data, __u32 size)
2624{
2625        struct btf_ext *btf_ext;
2626        int err;
2627
2628        err = btf_ext_parse_hdr(data, size);
2629        if (err)
2630                return libbpf_err_ptr(err);
2631
2632        btf_ext = calloc(1, sizeof(struct btf_ext));
2633        if (!btf_ext)
2634                return libbpf_err_ptr(-ENOMEM);
2635
2636        btf_ext->data_size = size;
2637        btf_ext->data = malloc(size);
2638        if (!btf_ext->data) {
2639                err = -ENOMEM;
2640                goto done;
2641        }
2642        memcpy(btf_ext->data, data, size);
2643
2644        if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, line_info_len)) {
2645                err = -EINVAL;
2646                goto done;
2647        }
2648
2649        err = btf_ext_setup_func_info(btf_ext);
2650        if (err)
2651                goto done;
2652
2653        err = btf_ext_setup_line_info(btf_ext);
2654        if (err)
2655                goto done;
2656
2657        if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, core_relo_len)) {
2658                err = -EINVAL;
2659                goto done;
2660        }
2661
2662        err = btf_ext_setup_core_relos(btf_ext);
2663        if (err)
2664                goto done;
2665
2666done:
2667        if (err) {
2668                btf_ext__free(btf_ext);
2669                return libbpf_err_ptr(err);
2670        }
2671
2672        return btf_ext;
2673}
2674
2675const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size)
2676{
2677        *size = btf_ext->data_size;
2678        return btf_ext->data;
2679}
2680
2681static int btf_ext_reloc_info(const struct btf *btf,
2682                              const struct btf_ext_info *ext_info,
2683                              const char *sec_name, __u32 insns_cnt,
2684                              void **info, __u32 *cnt)
2685{
2686        __u32 sec_hdrlen = sizeof(struct btf_ext_info_sec);
2687        __u32 i, record_size, existing_len, records_len;
2688        struct btf_ext_info_sec *sinfo;
2689        const char *info_sec_name;
2690        __u64 remain_len;
2691        void *data;
2692
2693        record_size = ext_info->rec_size;
2694        sinfo = ext_info->info;
2695        remain_len = ext_info->len;
2696        while (remain_len > 0) {
2697                records_len = sinfo->num_info * record_size;
2698                info_sec_name = btf__name_by_offset(btf, sinfo->sec_name_off);
2699                if (strcmp(info_sec_name, sec_name)) {
2700                        remain_len -= sec_hdrlen + records_len;
2701                        sinfo = (void *)sinfo + sec_hdrlen + records_len;
2702                        continue;
2703                }
2704
2705                existing_len = (*cnt) * record_size;
2706                data = realloc(*info, existing_len + records_len);
2707                if (!data)
2708                        return libbpf_err(-ENOMEM);
2709
2710                memcpy(data + existing_len, sinfo->data, records_len);
2711                /* adjust insn_off only, the rest data will be passed
2712                 * to the kernel.
2713                 */
2714                for (i = 0; i < sinfo->num_info; i++) {
2715                        __u32 *insn_off;
2716
2717                        insn_off = data + existing_len + (i * record_size);
2718                        *insn_off = *insn_off / sizeof(struct bpf_insn) + insns_cnt;
2719                }
2720                *info = data;
2721                *cnt += sinfo->num_info;
2722                return 0;
2723        }
2724
2725        return libbpf_err(-ENOENT);
2726}
2727
2728int btf_ext__reloc_func_info(const struct btf *btf,
2729                             const struct btf_ext *btf_ext,
2730                             const char *sec_name, __u32 insns_cnt,
2731                             void **func_info, __u32 *cnt)
2732{
2733        return btf_ext_reloc_info(btf, &btf_ext->func_info, sec_name,
2734                                  insns_cnt, func_info, cnt);
2735}
2736
2737int btf_ext__reloc_line_info(const struct btf *btf,
2738                             const struct btf_ext *btf_ext,
2739                             const char *sec_name, __u32 insns_cnt,
2740                             void **line_info, __u32 *cnt)
2741{
2742        return btf_ext_reloc_info(btf, &btf_ext->line_info, sec_name,
2743                                  insns_cnt, line_info, cnt);
2744}
2745
2746__u32 btf_ext__func_info_rec_size(const struct btf_ext *btf_ext)
2747{
2748        return btf_ext->func_info.rec_size;
2749}
2750
2751__u32 btf_ext__line_info_rec_size(const struct btf_ext *btf_ext)
2752{
2753        return btf_ext->line_info.rec_size;
2754}
2755
2756struct btf_dedup;
2757
2758static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
2759                                       const struct btf_dedup_opts *opts);
2760static void btf_dedup_free(struct btf_dedup *d);
2761static int btf_dedup_prep(struct btf_dedup *d);
2762static int btf_dedup_strings(struct btf_dedup *d);
2763static int btf_dedup_prim_types(struct btf_dedup *d);
2764static int btf_dedup_struct_types(struct btf_dedup *d);
2765static int btf_dedup_ref_types(struct btf_dedup *d);
2766static int btf_dedup_compact_types(struct btf_dedup *d);
2767static int btf_dedup_remap_types(struct btf_dedup *d);
2768
2769/*
2770 * Deduplicate BTF types and strings.
2771 *
2772 * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
2773 * section with all BTF type descriptors and string data. It overwrites that
2774 * memory in-place with deduplicated types and strings without any loss of
2775 * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
2776 * is provided, all the strings referenced from .BTF.ext section are honored
2777 * and updated to point to the right offsets after deduplication.
2778 *
2779 * If function returns with error, type/string data might be garbled and should
2780 * be discarded.
2781 *
2782 * More verbose and detailed description of both problem btf_dedup is solving,
2783 * as well as solution could be found at:
2784 * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
2785 *
2786 * Problem description and justification
2787 * =====================================
2788 *
2789 * BTF type information is typically emitted either as a result of conversion
2790 * from DWARF to BTF or directly by compiler. In both cases, each compilation
2791 * unit contains information about a subset of all the types that are used
2792 * in an application. These subsets are frequently overlapping and contain a lot
2793 * of duplicated information when later concatenated together into a single
2794 * binary. This algorithm ensures that each unique type is represented by single
2795 * BTF type descriptor, greatly reducing resulting size of BTF data.
2796 *
2797 * Compilation unit isolation and subsequent duplication of data is not the only
2798 * problem. The same type hierarchy (e.g., struct and all the type that struct
2799 * references) in different compilation units can be represented in BTF to
2800 * various degrees of completeness (or, rather, incompleteness) due to
2801 * struct/union forward declarations.
2802 *
2803 * Let's take a look at an example, that we'll use to better understand the
2804 * problem (and solution). Suppose we have two compilation units, each using
2805 * same `struct S`, but each of them having incomplete type information about
2806 * struct's fields:
2807 *
2808 * // CU #1:
2809 * struct S;
2810 * struct A {
2811 *      int a;
2812 *      struct A* self;
2813 *      struct S* parent;
2814 * };
2815 * struct B;
2816 * struct S {
2817 *      struct A* a_ptr;
2818 *      struct B* b_ptr;
2819 * };
2820 *
2821 * // CU #2:
2822 * struct S;
2823 * struct A;
2824 * struct B {
2825 *      int b;
2826 *      struct B* self;
2827 *      struct S* parent;
2828 * };
2829 * struct S {
2830 *      struct A* a_ptr;
2831 *      struct B* b_ptr;
2832 * };
2833 *
2834 * In case of CU #1, BTF data will know only that `struct B` exist (but no
2835 * more), but will know the complete type information about `struct A`. While
2836 * for CU #2, it will know full type information about `struct B`, but will
2837 * only know about forward declaration of `struct A` (in BTF terms, it will
2838 * have `BTF_KIND_FWD` type descriptor with name `B`).
2839 *
2840 * This compilation unit isolation means that it's possible that there is no
2841 * single CU with complete type information describing structs `S`, `A`, and
2842 * `B`. Also, we might get tons of duplicated and redundant type information.
2843 *
2844 * Additional complication we need to keep in mind comes from the fact that
2845 * types, in general, can form graphs containing cycles, not just DAGs.
2846 *
2847 * While algorithm does deduplication, it also merges and resolves type
2848 * information (unless disabled throught `struct btf_opts`), whenever possible.
2849 * E.g., in the example above with two compilation units having partial type
2850 * information for structs `A` and `B`, the output of algorithm will emit
2851 * a single copy of each BTF type that describes structs `A`, `B`, and `S`
2852 * (as well as type information for `int` and pointers), as if they were defined
2853 * in a single compilation unit as:
2854 *
2855 * struct A {
2856 *      int a;
2857 *      struct A* self;
2858 *      struct S* parent;
2859 * };
2860 * struct B {
2861 *      int b;
2862 *      struct B* self;
2863 *      struct S* parent;
2864 * };
2865 * struct S {
2866 *      struct A* a_ptr;
2867 *      struct B* b_ptr;
2868 * };
2869 *
2870 * Algorithm summary
2871 * =================
2872 *
2873 * Algorithm completes its work in 6 separate passes:
2874 *
2875 * 1. Strings deduplication.
2876 * 2. Primitive types deduplication (int, enum, fwd).
2877 * 3. Struct/union types deduplication.
2878 * 4. Reference types deduplication (pointers, typedefs, arrays, funcs, func
2879 *    protos, and const/volatile/restrict modifiers).
2880 * 5. Types compaction.
2881 * 6. Types remapping.
2882 *
2883 * Algorithm determines canonical type descriptor, which is a single
2884 * representative type for each truly unique type. This canonical type is the
2885 * one that will go into final deduplicated BTF type information. For
2886 * struct/unions, it is also the type that algorithm will merge additional type
2887 * information into (while resolving FWDs), as it discovers it from data in
2888 * other CUs. Each input BTF type eventually gets either mapped to itself, if
2889 * that type is canonical, or to some other type, if that type is equivalent
2890 * and was chosen as canonical representative. This mapping is stored in
2891 * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
2892 * FWD type got resolved to.
2893 *
2894 * To facilitate fast discovery of canonical types, we also maintain canonical
2895 * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
2896 * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
2897 * that match that signature. With sufficiently good choice of type signature
2898 * hashing function, we can limit number of canonical types for each unique type
2899 * signature to a very small number, allowing to find canonical type for any
2900 * duplicated type very quickly.
2901 *
2902 * Struct/union deduplication is the most critical part and algorithm for
2903 * deduplicating structs/unions is described in greater details in comments for
2904 * `btf_dedup_is_equiv` function.
2905 */
2906int btf__dedup(struct btf *btf, struct btf_ext *btf_ext,
2907               const struct btf_dedup_opts *opts)
2908{
2909        struct btf_dedup *d = btf_dedup_new(btf, btf_ext, opts);
2910        int err;
2911
2912        if (IS_ERR(d)) {
2913                pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
2914                return libbpf_err(-EINVAL);
2915        }
2916
2917        if (btf_ensure_modifiable(btf))
2918                return libbpf_err(-ENOMEM);
2919
2920        err = btf_dedup_prep(d);
2921        if (err) {
2922                pr_debug("btf_dedup_prep failed:%d\n", err);
2923                goto done;
2924        }
2925        err = btf_dedup_strings(d);
2926        if (err < 0) {
2927                pr_debug("btf_dedup_strings failed:%d\n", err);
2928                goto done;
2929        }
2930        err = btf_dedup_prim_types(d);
2931        if (err < 0) {
2932                pr_debug("btf_dedup_prim_types failed:%d\n", err);
2933                goto done;
2934        }
2935        err = btf_dedup_struct_types(d);
2936        if (err < 0) {
2937                pr_debug("btf_dedup_struct_types failed:%d\n", err);
2938                goto done;
2939        }
2940        err = btf_dedup_ref_types(d);
2941        if (err < 0) {
2942                pr_debug("btf_dedup_ref_types failed:%d\n", err);
2943                goto done;
2944        }
2945        err = btf_dedup_compact_types(d);
2946        if (err < 0) {
2947                pr_debug("btf_dedup_compact_types failed:%d\n", err);
2948                goto done;
2949        }
2950        err = btf_dedup_remap_types(d);
2951        if (err < 0) {
2952                pr_debug("btf_dedup_remap_types failed:%d\n", err);
2953                goto done;
2954        }
2955
2956done:
2957        btf_dedup_free(d);
2958        return libbpf_err(err);
2959}
2960
2961#define BTF_UNPROCESSED_ID ((__u32)-1)
2962#define BTF_IN_PROGRESS_ID ((__u32)-2)
2963
2964struct btf_dedup {
2965        /* .BTF section to be deduped in-place */
2966        struct btf *btf;
2967        /*
2968         * Optional .BTF.ext section. When provided, any strings referenced
2969         * from it will be taken into account when deduping strings
2970         */
2971        struct btf_ext *btf_ext;
2972        /*
2973         * This is a map from any type's signature hash to a list of possible
2974         * canonical representative type candidates. Hash collisions are
2975         * ignored, so even types of various kinds can share same list of
2976         * candidates, which is fine because we rely on subsequent
2977         * btf_xxx_equal() checks to authoritatively verify type equality.
2978         */
2979        struct hashmap *dedup_table;
2980        /* Canonical types map */
2981        __u32 *map;
2982        /* Hypothetical mapping, used during type graph equivalence checks */
2983        __u32 *hypot_map;
2984        __u32 *hypot_list;
2985        size_t hypot_cnt;
2986        size_t hypot_cap;
2987        /* Whether hypothetical mapping, if successful, would need to adjust
2988         * already canonicalized types (due to a new forward declaration to
2989         * concrete type resolution). In such case, during split BTF dedup
2990         * candidate type would still be considered as different, because base
2991         * BTF is considered to be immutable.
2992         */
2993        bool hypot_adjust_canon;
2994        /* Various option modifying behavior of algorithm */
2995        struct btf_dedup_opts opts;
2996        /* temporary strings deduplication state */
2997        struct strset *strs_set;
2998};
2999
3000static long hash_combine(long h, long value)
3001{
3002        return h * 31 + value;
3003}
3004
3005#define for_each_dedup_cand(d, node, hash) \
3006        hashmap__for_each_key_entry(d->dedup_table, node, (void *)hash)
3007
3008static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
3009{
3010        return hashmap__append(d->dedup_table,
3011                               (void *)hash, (void *)(long)type_id);
3012}
3013
3014static int btf_dedup_hypot_map_add(struct btf_dedup *d,
3015                                   __u32 from_id, __u32 to_id)
3016{
3017        if (d->hypot_cnt == d->hypot_cap) {
3018                __u32 *new_list;
3019
3020                d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
3021                new_list = libbpf_reallocarray(d->hypot_list, d->hypot_cap, sizeof(__u32));
3022                if (!new_list)
3023                        return -ENOMEM;
3024                d->hypot_list = new_list;
3025        }
3026        d->hypot_list[d->hypot_cnt++] = from_id;
3027        d->hypot_map[from_id] = to_id;
3028        return 0;
3029}
3030
3031static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
3032{
3033        int i;
3034
3035        for (i = 0; i < d->hypot_cnt; i++)
3036                d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
3037        d->hypot_cnt = 0;
3038        d->hypot_adjust_canon = false;
3039}
3040
3041static void btf_dedup_free(struct btf_dedup *d)
3042{
3043        hashmap__free(d->dedup_table);
3044        d->dedup_table = NULL;
3045
3046        free(d->map);
3047        d->map = NULL;
3048
3049        free(d->hypot_map);
3050        d->hypot_map = NULL;
3051
3052        free(d->hypot_list);
3053        d->hypot_list = NULL;
3054
3055        free(d);
3056}
3057
3058static size_t btf_dedup_identity_hash_fn(const void *key, void *ctx)
3059{
3060        return (size_t)key;
3061}
3062
3063static size_t btf_dedup_collision_hash_fn(const void *key, void *ctx)
3064{
3065        return 0;
3066}
3067
3068static bool btf_dedup_equal_fn(const void *k1, const void *k2, void *ctx)
3069{
3070        return k1 == k2;
3071}
3072
3073static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
3074                                       const struct btf_dedup_opts *opts)
3075{
3076        struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
3077        hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
3078        int i, err = 0, type_cnt;
3079
3080        if (!d)
3081                return ERR_PTR(-ENOMEM);
3082
3083        d->opts.dont_resolve_fwds = opts && opts->dont_resolve_fwds;
3084        /* dedup_table_size is now used only to force collisions in tests */
3085        if (opts && opts->dedup_table_size == 1)
3086                hash_fn = btf_dedup_collision_hash_fn;
3087
3088        d->btf = btf;
3089        d->btf_ext = btf_ext;
3090
3091        d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
3092        if (IS_ERR(d->dedup_table)) {
3093                err = PTR_ERR(d->dedup_table);
3094                d->dedup_table = NULL;
3095                goto done;
3096        }
3097
3098        type_cnt = btf__get_nr_types(btf) + 1;
3099        d->map = malloc(sizeof(__u32) * type_cnt);
3100        if (!d->map) {
3101                err = -ENOMEM;
3102                goto done;
3103        }
3104        /* special BTF "void" type is made canonical immediately */
3105        d->map[0] = 0;
3106        for (i = 1; i < type_cnt; i++) {
3107                struct btf_type *t = btf_type_by_id(d->btf, i);
3108
3109                /* VAR and DATASEC are never deduped and are self-canonical */
3110                if (btf_is_var(t) || btf_is_datasec(t))
3111                        d->map[i] = i;
3112                else
3113                        d->map[i] = BTF_UNPROCESSED_ID;
3114        }
3115
3116        d->hypot_map = malloc(sizeof(__u32) * type_cnt);
3117        if (!d->hypot_map) {
3118                err = -ENOMEM;
3119                goto done;
3120        }
3121        for (i = 0; i < type_cnt; i++)
3122                d->hypot_map[i] = BTF_UNPROCESSED_ID;
3123
3124done:
3125        if (err) {
3126                btf_dedup_free(d);
3127                return ERR_PTR(err);
3128        }
3129
3130        return d;
3131}
3132
3133/*
3134 * Iterate over all possible places in .BTF and .BTF.ext that can reference
3135 * string and pass pointer to it to a provided callback `fn`.
3136 */
3137static int btf_for_each_str_off(struct btf_dedup *d, str_off_visit_fn fn, void *ctx)
3138{
3139        int i, r;
3140
3141        for (i = 0; i < d->btf->nr_types; i++) {
3142                struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
3143
3144                r = btf_type_visit_str_offs(t, fn, ctx);
3145                if (r)
3146                        return r;
3147        }
3148
3149        if (!d->btf_ext)
3150                return 0;
3151
3152        r = btf_ext_visit_str_offs(d->btf_ext, fn, ctx);
3153        if (r)
3154                return r;
3155
3156        return 0;
3157}
3158
3159static int strs_dedup_remap_str_off(__u32 *str_off_ptr, void *ctx)
3160{
3161        struct btf_dedup *d = ctx;
3162        __u32 str_off = *str_off_ptr;
3163        const char *s;
3164        int off, err;
3165
3166        /* don't touch empty string or string in main BTF */
3167        if (str_off == 0 || str_off < d->btf->start_str_off)
3168                return 0;
3169
3170        s = btf__str_by_offset(d->btf, str_off);
3171        if (d->btf->base_btf) {
3172                err = btf__find_str(d->btf->base_btf, s);
3173                if (err >= 0) {
3174                        *str_off_ptr = err;
3175                        return 0;
3176                }
3177                if (err != -ENOENT)
3178                        return err;
3179        }
3180
3181        off = strset__add_str(d->strs_set, s);
3182        if (off < 0)
3183                return off;
3184
3185        *str_off_ptr = d->btf->start_str_off + off;
3186        return 0;
3187}
3188
3189/*
3190 * Dedup string and filter out those that are not referenced from either .BTF
3191 * or .BTF.ext (if provided) sections.
3192 *
3193 * This is done by building index of all strings in BTF's string section,
3194 * then iterating over all entities that can reference strings (e.g., type
3195 * names, struct field names, .BTF.ext line info, etc) and marking corresponding
3196 * strings as used. After that all used strings are deduped and compacted into
3197 * sequential blob of memory and new offsets are calculated. Then all the string
3198 * references are iterated again and rewritten using new offsets.
3199 */
3200static int btf_dedup_strings(struct btf_dedup *d)
3201{
3202        int err;
3203
3204        if (d->btf->strs_deduped)
3205                return 0;
3206
3207        d->strs_set = strset__new(BTF_MAX_STR_OFFSET, NULL, 0);
3208        if (IS_ERR(d->strs_set)) {
3209                err = PTR_ERR(d->strs_set);
3210                goto err_out;
3211        }
3212
3213        if (!d->btf->base_btf) {
3214                /* insert empty string; we won't be looking it up during strings
3215                 * dedup, but it's good to have it for generic BTF string lookups
3216                 */
3217                err = strset__add_str(d->strs_set, "");
3218                if (err < 0)
3219                        goto err_out;
3220        }
3221
3222        /* remap string offsets */
3223        err = btf_for_each_str_off(d, strs_dedup_remap_str_off, d);
3224        if (err)
3225                goto err_out;
3226
3227        /* replace BTF string data and hash with deduped ones */
3228        strset__free(d->btf->strs_set);
3229        d->btf->hdr->str_len = strset__data_size(d->strs_set);
3230        d->btf->strs_set = d->strs_set;
3231        d->strs_set = NULL;
3232        d->btf->strs_deduped = true;
3233        return 0;
3234
3235err_out:
3236        strset__free(d->strs_set);
3237        d->strs_set = NULL;
3238
3239        return err;
3240}
3241
3242static long btf_hash_common(struct btf_type *t)
3243{
3244        long h;
3245
3246        h = hash_combine(0, t->name_off);
3247        h = hash_combine(h, t->info);
3248        h = hash_combine(h, t->size);
3249        return h;
3250}
3251
3252static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
3253{
3254        return t1->name_off == t2->name_off &&
3255               t1->info == t2->info &&
3256               t1->size == t2->size;
3257}
3258
3259/* Calculate type signature hash of INT. */
3260static long btf_hash_int(struct btf_type *t)
3261{
3262        __u32 info = *(__u32 *)(t + 1);
3263        long h;
3264
3265        h = btf_hash_common(t);
3266        h = hash_combine(h, info);
3267        return h;
3268}
3269
3270/* Check structural equality of two INTs. */
3271static bool btf_equal_int(struct btf_type *t1, struct btf_type *t2)
3272{
3273        __u32 info1, info2;
3274
3275        if (!btf_equal_common(t1, t2))
3276                return false;
3277        info1 = *(__u32 *)(t1 + 1);
3278        info2 = *(__u32 *)(t2 + 1);
3279        return info1 == info2;
3280}
3281
3282/* Calculate type signature hash of ENUM. */
3283static long btf_hash_enum(struct btf_type *t)
3284{
3285        long h;
3286
3287        /* don't hash vlen and enum members to support enum fwd resolving */
3288        h = hash_combine(0, t->name_off);
3289        h = hash_combine(h, t->info & ~0xffff);
3290        h = hash_combine(h, t->size);
3291        return h;
3292}
3293
3294/* Check structural equality of two ENUMs. */
3295static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
3296{
3297        const struct btf_enum *m1, *m2;
3298        __u16 vlen;
3299        int i;
3300
3301        if (!btf_equal_common(t1, t2))
3302                return false;
3303
3304        vlen = btf_vlen(t1);
3305        m1 = btf_enum(t1);
3306        m2 = btf_enum(t2);
3307        for (i = 0; i < vlen; i++) {
3308                if (m1->name_off != m2->name_off || m1->val != m2->val)
3309                        return false;
3310                m1++;
3311                m2++;
3312        }
3313        return true;
3314}
3315
3316static inline bool btf_is_enum_fwd(struct btf_type *t)
3317{
3318        return btf_is_enum(t) && btf_vlen(t) == 0;
3319}
3320
3321static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
3322{
3323        if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3324                return btf_equal_enum(t1, t2);
3325        /* ignore vlen when comparing */
3326        return t1->name_off == t2->name_off &&
3327               (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
3328               t1->size == t2->size;
3329}
3330
3331/*
3332 * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
3333 * as referenced type IDs equivalence is established separately during type
3334 * graph equivalence check algorithm.
3335 */
3336static long btf_hash_struct(struct btf_type *t)
3337{
3338        const struct btf_member *member = btf_members(t);
3339        __u32 vlen = btf_vlen(t);
3340        long h = btf_hash_common(t);
3341        int i;
3342
3343        for (i = 0; i < vlen; i++) {
3344                h = hash_combine(h, member->name_off);
3345                h = hash_combine(h, member->offset);
3346                /* no hashing of referenced type ID, it can be unresolved yet */
3347                member++;
3348        }
3349        return h;
3350}
3351
3352/*
3353 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3354 * IDs. This check is performed during type graph equivalence check and
3355 * referenced types equivalence is checked separately.
3356 */
3357static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
3358{
3359        const struct btf_member *m1, *m2;
3360        __u16 vlen;
3361        int i;
3362
3363        if (!btf_equal_common(t1, t2))
3364                return false;
3365
3366        vlen = btf_vlen(t1);
3367        m1 = btf_members(t1);
3368        m2 = btf_members(t2);
3369        for (i = 0; i < vlen; i++) {
3370                if (m1->name_off != m2->name_off || m1->offset != m2->offset)
3371                        return false;
3372                m1++;
3373                m2++;
3374        }
3375        return true;
3376}
3377
3378/*
3379 * Calculate type signature hash of ARRAY, including referenced type IDs,
3380 * under assumption that they were already resolved to canonical type IDs and
3381 * are not going to change.
3382 */
3383static long btf_hash_array(struct btf_type *t)
3384{
3385        const struct btf_array *info = btf_array(t);
3386        long h = btf_hash_common(t);
3387
3388        h = hash_combine(h, info->type);
3389        h = hash_combine(h, info->index_type);
3390        h = hash_combine(h, info->nelems);
3391        return h;
3392}
3393
3394/*
3395 * Check exact equality of two ARRAYs, taking into account referenced
3396 * type IDs, under assumption that they were already resolved to canonical
3397 * type IDs and are not going to change.
3398 * This function is called during reference types deduplication to compare
3399 * ARRAY to potential canonical representative.
3400 */
3401static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
3402{
3403        const struct btf_array *info1, *info2;
3404
3405        if (!btf_equal_common(t1, t2))
3406                return false;
3407
3408        info1 = btf_array(t1);
3409        info2 = btf_array(t2);
3410        return info1->type == info2->type &&
3411               info1->index_type == info2->index_type &&
3412               info1->nelems == info2->nelems;
3413}
3414
3415/*
3416 * Check structural compatibility of two ARRAYs, ignoring referenced type
3417 * IDs. This check is performed during type graph equivalence check and
3418 * referenced types equivalence is checked separately.
3419 */
3420static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
3421{
3422        if (!btf_equal_common(t1, t2))
3423                return false;
3424
3425        return btf_array(t1)->nelems == btf_array(t2)->nelems;
3426}
3427
3428/*
3429 * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
3430 * under assumption that they were already resolved to canonical type IDs and
3431 * are not going to change.
3432 */
3433static long btf_hash_fnproto(struct btf_type *t)
3434{
3435        const struct btf_param *member = btf_params(t);
3436        __u16 vlen = btf_vlen(t);
3437        long h = btf_hash_common(t);
3438        int i;
3439
3440        for (i = 0; i < vlen; i++) {
3441                h = hash_combine(h, member->name_off);
3442                h = hash_combine(h, member->type);
3443                member++;
3444        }
3445        return h;
3446}
3447
3448/*
3449 * Check exact equality of two FUNC_PROTOs, taking into account referenced
3450 * type IDs, under assumption that they were already resolved to canonical
3451 * type IDs and are not going to change.
3452 * This function is called during reference types deduplication to compare
3453 * FUNC_PROTO to potential canonical representative.
3454 */
3455static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
3456{
3457        const struct btf_param *m1, *m2;
3458        __u16 vlen;
3459        int i;
3460
3461        if (!btf_equal_common(t1, t2))
3462                return false;
3463
3464        vlen = btf_vlen(t1);
3465        m1 = btf_params(t1);
3466        m2 = btf_params(t2);
3467        for (i = 0; i < vlen; i++) {
3468                if (m1->name_off != m2->name_off || m1->type != m2->type)
3469                        return false;
3470                m1++;
3471                m2++;
3472        }
3473        return true;
3474}
3475
3476/*
3477 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3478 * IDs. This check is performed during type graph equivalence check and
3479 * referenced types equivalence is checked separately.
3480 */
3481static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
3482{
3483        const struct btf_param *m1, *m2;
3484        __u16 vlen;
3485        int i;
3486
3487        /* skip return type ID */
3488        if (t1->name_off != t2->name_off || t1->info != t2->info)
3489                return false;
3490
3491        vlen = btf_vlen(t1);
3492        m1 = btf_params(t1);
3493        m2 = btf_params(t2);
3494        for (i = 0; i < vlen; i++) {
3495                if (m1->name_off != m2->name_off)
3496                        return false;
3497                m1++;
3498                m2++;
3499        }
3500        return true;
3501}
3502
3503/* Prepare split BTF for deduplication by calculating hashes of base BTF's
3504 * types and initializing the rest of the state (canonical type mapping) for
3505 * the fixed base BTF part.
3506 */
3507static int btf_dedup_prep(struct btf_dedup *d)
3508{
3509        struct btf_type *t;
3510        int type_id;
3511        long h;
3512
3513        if (!d->btf->base_btf)
3514                return 0;
3515
3516        for (type_id = 1; type_id < d->btf->start_id; type_id++) {
3517                t = btf_type_by_id(d->btf, type_id);
3518
3519                /* all base BTF types are self-canonical by definition */
3520                d->map[type_id] = type_id;
3521
3522                switch (btf_kind(t)) {
3523                case BTF_KIND_VAR:
3524                case BTF_KIND_DATASEC:
3525                        /* VAR and DATASEC are never hash/deduplicated */
3526                        continue;
3527                case BTF_KIND_CONST:
3528                case BTF_KIND_VOLATILE:
3529                case BTF_KIND_RESTRICT:
3530                case BTF_KIND_PTR:
3531                case BTF_KIND_FWD:
3532                case BTF_KIND_TYPEDEF:
3533                case BTF_KIND_FUNC:
3534                case BTF_KIND_FLOAT:
3535                        h = btf_hash_common(t);
3536                        break;
3537                case BTF_KIND_INT:
3538                        h = btf_hash_int(t);
3539                        break;
3540                case BTF_KIND_ENUM:
3541                        h = btf_hash_enum(t);
3542                        break;
3543                case BTF_KIND_STRUCT:
3544                case BTF_KIND_UNION:
3545                        h = btf_hash_struct(t);
3546                        break;
3547                case BTF_KIND_ARRAY:
3548                        h = btf_hash_array(t);
3549                        break;
3550                case BTF_KIND_FUNC_PROTO:
3551                        h = btf_hash_fnproto(t);
3552                        break;
3553                default:
3554                        pr_debug("unknown kind %d for type [%d]\n", btf_kind(t), type_id);
3555                        return -EINVAL;
3556                }
3557                if (btf_dedup_table_add(d, h, type_id))
3558                        return -ENOMEM;
3559        }
3560
3561        return 0;
3562}
3563
3564/*
3565 * Deduplicate primitive types, that can't reference other types, by calculating
3566 * their type signature hash and comparing them with any possible canonical
3567 * candidate. If no canonical candidate matches, type itself is marked as
3568 * canonical and is added into `btf_dedup->dedup_table` as another candidate.
3569 */
3570static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
3571{
3572        struct btf_type *t = btf_type_by_id(d->btf, type_id);
3573        struct hashmap_entry *hash_entry;
3574        struct btf_type *cand;
3575        /* if we don't find equivalent type, then we are canonical */
3576        __u32 new_id = type_id;
3577        __u32 cand_id;
3578        long h;
3579
3580        switch (btf_kind(t)) {
3581        case BTF_KIND_CONST:
3582        case BTF_KIND_VOLATILE:
3583        case BTF_KIND_RESTRICT:
3584        case BTF_KIND_PTR:
3585        case BTF_KIND_TYPEDEF:
3586        case BTF_KIND_ARRAY:
3587        case BTF_KIND_STRUCT:
3588        case BTF_KIND_UNION:
3589        case BTF_KIND_FUNC:
3590        case BTF_KIND_FUNC_PROTO:
3591        case BTF_KIND_VAR:
3592        case BTF_KIND_DATASEC:
3593                return 0;
3594
3595        case BTF_KIND_INT:
3596                h = btf_hash_int(t);
3597                for_each_dedup_cand(d, hash_entry, h) {
3598                        cand_id = (__u32)(long)hash_entry->value;
3599                        cand = btf_type_by_id(d->btf, cand_id);
3600                        if (btf_equal_int(t, cand)) {
3601                                new_id = cand_id;
3602                                break;
3603                        }
3604                }
3605                break;
3606
3607        case BTF_KIND_ENUM:
3608                h = btf_hash_enum(t);
3609                for_each_dedup_cand(d, hash_entry, h) {
3610                        cand_id = (__u32)(long)hash_entry->value;
3611                        cand = btf_type_by_id(d->btf, cand_id);
3612                        if (btf_equal_enum(t, cand)) {
3613                                new_id = cand_id;
3614                                break;
3615                        }
3616                        if (d->opts.dont_resolve_fwds)
3617                                continue;
3618                        if (btf_compat_enum(t, cand)) {
3619                                if (btf_is_enum_fwd(t)) {
3620                                        /* resolve fwd to full enum */
3621                                        new_id = cand_id;
3622                                        break;
3623                                }
3624                                /* resolve canonical enum fwd to full enum */
3625                                d->map[cand_id] = type_id;
3626                        }
3627                }
3628                break;
3629
3630        case BTF_KIND_FWD:
3631        case BTF_KIND_FLOAT:
3632                h = btf_hash_common(t);
3633                for_each_dedup_cand(d, hash_entry, h) {
3634                        cand_id = (__u32)(long)hash_entry->value;
3635                        cand = btf_type_by_id(d->btf, cand_id);
3636                        if (btf_equal_common(t, cand)) {
3637                                new_id = cand_id;
3638                                break;
3639                        }
3640                }
3641                break;
3642
3643        default:
3644                return -EINVAL;
3645        }
3646
3647        d->map[type_id] = new_id;
3648        if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
3649                return -ENOMEM;
3650
3651        return 0;
3652}
3653
3654static int btf_dedup_prim_types(struct btf_dedup *d)
3655{
3656        int i, err;
3657
3658        for (i = 0; i < d->btf->nr_types; i++) {
3659                err = btf_dedup_prim_type(d, d->btf->start_id + i);
3660                if (err)
3661                        return err;
3662        }
3663        return 0;
3664}
3665
3666/*
3667 * Check whether type is already mapped into canonical one (could be to itself).
3668 */
3669static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
3670{
3671        return d->map[type_id] <= BTF_MAX_NR_TYPES;
3672}
3673
3674/*
3675 * Resolve type ID into its canonical type ID, if any; otherwise return original
3676 * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
3677 * STRUCT/UNION link and resolve it into canonical type ID as well.
3678 */
3679static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
3680{
3681        while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3682                type_id = d->map[type_id];
3683        return type_id;
3684}
3685
3686/*
3687 * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
3688 * type ID.
3689 */
3690static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
3691{
3692        __u32 orig_type_id = type_id;
3693
3694        if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3695                return type_id;
3696
3697        while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3698                type_id = d->map[type_id];
3699
3700        if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3701                return type_id;
3702
3703        return orig_type_id;
3704}
3705
3706
3707static inline __u16 btf_fwd_kind(struct btf_type *t)
3708{
3709        return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
3710}
3711
3712/* Check if given two types are identical ARRAY definitions */
3713static int btf_dedup_identical_arrays(struct btf_dedup *d, __u32 id1, __u32 id2)
3714{
3715        struct btf_type *t1, *t2;
3716
3717        t1 = btf_type_by_id(d->btf, id1);
3718        t2 = btf_type_by_id(d->btf, id2);
3719        if (!btf_is_array(t1) || !btf_is_array(t2))
3720                return 0;
3721
3722        return btf_equal_array(t1, t2);
3723}
3724
3725/*
3726 * Check equivalence of BTF type graph formed by candidate struct/union (we'll
3727 * call it "candidate graph" in this description for brevity) to a type graph
3728 * formed by (potential) canonical struct/union ("canonical graph" for brevity
3729 * here, though keep in mind that not all types in canonical graph are
3730 * necessarily canonical representatives themselves, some of them might be
3731 * duplicates or its uniqueness might not have been established yet).
3732 * Returns:
3733 *  - >0, if type graphs are equivalent;
3734 *  -  0, if not equivalent;
3735 *  - <0, on error.
3736 *
3737 * Algorithm performs side-by-side DFS traversal of both type graphs and checks
3738 * equivalence of BTF types at each step. If at any point BTF types in candidate
3739 * and canonical graphs are not compatible structurally, whole graphs are
3740 * incompatible. If types are structurally equivalent (i.e., all information
3741 * except referenced type IDs is exactly the same), a mapping from `canon_id` to
3742 * a `cand_id` is recored in hypothetical mapping (`btf_dedup->hypot_map`).
3743 * If a type references other types, then those referenced types are checked
3744 * for equivalence recursively.
3745 *
3746 * During DFS traversal, if we find that for current `canon_id` type we
3747 * already have some mapping in hypothetical map, we check for two possible
3748 * situations:
3749 *   - `canon_id` is mapped to exactly the same type as `cand_id`. This will
3750 *     happen when type graphs have cycles. In this case we assume those two
3751 *     types are equivalent.
3752 *   - `canon_id` is mapped to different type. This is contradiction in our
3753 *     hypothetical mapping, because same graph in canonical graph corresponds
3754 *     to two different types in candidate graph, which for equivalent type
3755 *     graphs shouldn't happen. This condition terminates equivalence check
3756 *     with negative result.
3757 *
3758 * If type graphs traversal exhausts types to check and find no contradiction,
3759 * then type graphs are equivalent.
3760 *
3761 * When checking types for equivalence, there is one special case: FWD types.
3762 * If FWD type resolution is allowed and one of the types (either from canonical
3763 * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
3764 * flag) and their names match, hypothetical mapping is updated to point from
3765 * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
3766 * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
3767 *
3768 * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
3769 * if there are two exactly named (or anonymous) structs/unions that are
3770 * compatible structurally, one of which has FWD field, while other is concrete
3771 * STRUCT/UNION, but according to C sources they are different structs/unions
3772 * that are referencing different types with the same name. This is extremely
3773 * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
3774 * this logic is causing problems.
3775 *
3776 * Doing FWD resolution means that both candidate and/or canonical graphs can
3777 * consists of portions of the graph that come from multiple compilation units.
3778 * This is due to the fact that types within single compilation unit are always
3779 * deduplicated and FWDs are already resolved, if referenced struct/union
3780 * definiton is available. So, if we had unresolved FWD and found corresponding
3781 * STRUCT/UNION, they will be from different compilation units. This
3782 * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
3783 * type graph will likely have at least two different BTF types that describe
3784 * same type (e.g., most probably there will be two different BTF types for the
3785 * same 'int' primitive type) and could even have "overlapping" parts of type
3786 * graph that describe same subset of types.
3787 *
3788 * This in turn means that our assumption that each type in canonical graph
3789 * must correspond to exactly one type in candidate graph might not hold
3790 * anymore and will make it harder to detect contradictions using hypothetical
3791 * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
3792 * resolution only in canonical graph. FWDs in candidate graphs are never
3793 * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
3794 * that can occur:
3795 *   - Both types in canonical and candidate graphs are FWDs. If they are
3796 *     structurally equivalent, then they can either be both resolved to the
3797 *     same STRUCT/UNION or not resolved at all. In both cases they are
3798 *     equivalent and there is no need to resolve FWD on candidate side.
3799 *   - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
3800 *     so nothing to resolve as well, algorithm will check equivalence anyway.
3801 *   - Type in canonical graph is FWD, while type in candidate is concrete
3802 *     STRUCT/UNION. In this case candidate graph comes from single compilation
3803 *     unit, so there is exactly one BTF type for each unique C type. After
3804 *     resolving FWD into STRUCT/UNION, there might be more than one BTF type
3805 *     in canonical graph mapping to single BTF type in candidate graph, but
3806 *     because hypothetical mapping maps from canonical to candidate types, it's
3807 *     alright, and we still maintain the property of having single `canon_id`
3808 *     mapping to single `cand_id` (there could be two different `canon_id`
3809 *     mapped to the same `cand_id`, but it's not contradictory).
3810 *   - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
3811 *     graph is FWD. In this case we are just going to check compatibility of
3812 *     STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
3813 *     assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
3814 *     a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
3815 *     turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
3816 *     canonical graph.
3817 */
3818static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
3819                              __u32 canon_id)
3820{
3821        struct btf_type *cand_type;
3822        struct btf_type *canon_type;
3823        __u32 hypot_type_id;
3824        __u16 cand_kind;
3825        __u16 canon_kind;
3826        int i, eq;
3827
3828        /* if both resolve to the same canonical, they must be equivalent */
3829        if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
3830                return 1;
3831
3832        canon_id = resolve_fwd_id(d, canon_id);
3833
3834        hypot_type_id = d->hypot_map[canon_id];
3835        if (hypot_type_id <= BTF_MAX_NR_TYPES) {
3836                /* In some cases compiler will generate different DWARF types
3837                 * for *identical* array type definitions and use them for
3838                 * different fields within the *same* struct. This breaks type
3839                 * equivalence check, which makes an assumption that candidate
3840                 * types sub-graph has a consistent and deduped-by-compiler
3841                 * types within a single CU. So work around that by explicitly
3842                 * allowing identical array types here.
3843                 */
3844                return hypot_type_id == cand_id ||
3845                       btf_dedup_identical_arrays(d, hypot_type_id, cand_id);
3846        }
3847
3848        if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
3849                return -ENOMEM;
3850
3851        cand_type = btf_type_by_id(d->btf, cand_id);
3852        canon_type = btf_type_by_id(d->btf, canon_id);
3853        cand_kind = btf_kind(cand_type);
3854        canon_kind = btf_kind(canon_type);
3855
3856        if (cand_type->name_off != canon_type->name_off)
3857                return 0;
3858
3859        /* FWD <--> STRUCT/UNION equivalence check, if enabled */
3860        if (!d->opts.dont_resolve_fwds
3861            && (cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
3862            && cand_kind != canon_kind) {
3863                __u16 real_kind;
3864                __u16 fwd_kind;
3865
3866                if (cand_kind == BTF_KIND_FWD) {
3867                        real_kind = canon_kind;
3868                        fwd_kind = btf_fwd_kind(cand_type);
3869                } else {
3870                        real_kind = cand_kind;
3871                        fwd_kind = btf_fwd_kind(canon_type);
3872                        /* we'd need to resolve base FWD to STRUCT/UNION */
3873                        if (fwd_kind == real_kind && canon_id < d->btf->start_id)
3874                                d->hypot_adjust_canon = true;
3875                }
3876                return fwd_kind == real_kind;
3877        }
3878
3879        if (cand_kind != canon_kind)
3880                return 0;
3881
3882        switch (cand_kind) {
3883        case BTF_KIND_INT:
3884                return btf_equal_int(cand_type, canon_type);
3885
3886        case BTF_KIND_ENUM:
3887                if (d->opts.dont_resolve_fwds)
3888                        return btf_equal_enum(cand_type, canon_type);
3889                else
3890                        return btf_compat_enum(cand_type, canon_type);
3891
3892        case BTF_KIND_FWD:
3893        case BTF_KIND_FLOAT:
3894                return btf_equal_common(cand_type, canon_type);
3895
3896        case BTF_KIND_CONST:
3897        case BTF_KIND_VOLATILE:
3898        case BTF_KIND_RESTRICT:
3899        case BTF_KIND_PTR:
3900        case BTF_KIND_TYPEDEF:
3901        case BTF_KIND_FUNC:
3902                if (cand_type->info != canon_type->info)
3903                        return 0;
3904                return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
3905
3906        case BTF_KIND_ARRAY: {
3907                const struct btf_array *cand_arr, *canon_arr;
3908
3909                if (!btf_compat_array(cand_type, canon_type))
3910                        return 0;
3911                cand_arr = btf_array(cand_type);
3912                canon_arr = btf_array(canon_type);
3913                eq = btf_dedup_is_equiv(d, cand_arr->index_type, canon_arr->index_type);
3914                if (eq <= 0)
3915                        return eq;
3916                return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
3917        }
3918
3919        case BTF_KIND_STRUCT:
3920        case BTF_KIND_UNION: {
3921                const struct btf_member *cand_m, *canon_m;
3922                __u16 vlen;
3923
3924                if (!btf_shallow_equal_struct(cand_type, canon_type))
3925                        return 0;
3926                vlen = btf_vlen(cand_type);
3927                cand_m = btf_members(cand_type);
3928                canon_m = btf_members(canon_type);
3929                for (i = 0; i < vlen; i++) {
3930                        eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
3931                        if (eq <= 0)
3932                                return eq;
3933                        cand_m++;
3934                        canon_m++;
3935                }
3936
3937                return 1;
3938        }
3939
3940        case BTF_KIND_FUNC_PROTO: {
3941                const struct btf_param *cand_p, *canon_p;
3942                __u16 vlen;
3943
3944                if (!btf_compat_fnproto(cand_type, canon_type))
3945                        return 0;
3946                eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
3947                if (eq <= 0)
3948                        return eq;
3949                vlen = btf_vlen(cand_type);
3950                cand_p = btf_params(cand_type);
3951                canon_p = btf_params(canon_type);
3952                for (i = 0; i < vlen; i++) {
3953                        eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
3954                        if (eq <= 0)
3955                                return eq;
3956                        cand_p++;
3957                        canon_p++;
3958                }
3959                return 1;
3960        }
3961
3962        default:
3963                return -EINVAL;
3964        }
3965        return 0;
3966}
3967
3968/*
3969 * Use hypothetical mapping, produced by successful type graph equivalence
3970 * check, to augment existing struct/union canonical mapping, where possible.
3971 *
3972 * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
3973 * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
3974 * it doesn't matter if FWD type was part of canonical graph or candidate one,
3975 * we are recording the mapping anyway. As opposed to carefulness required
3976 * for struct/union correspondence mapping (described below), for FWD resolution
3977 * it's not important, as by the time that FWD type (reference type) will be
3978 * deduplicated all structs/unions will be deduped already anyway.
3979 *
3980 * Recording STRUCT/UNION mapping is purely a performance optimization and is
3981 * not required for correctness. It needs to be done carefully to ensure that
3982 * struct/union from candidate's type graph is not mapped into corresponding
3983 * struct/union from canonical type graph that itself hasn't been resolved into
3984 * canonical representative. The only guarantee we have is that canonical
3985 * struct/union was determined as canonical and that won't change. But any
3986 * types referenced through that struct/union fields could have been not yet
3987 * resolved, so in case like that it's too early to establish any kind of
3988 * correspondence between structs/unions.
3989 *
3990 * No canonical correspondence is derived for primitive types (they are already
3991 * deduplicated completely already anyway) or reference types (they rely on
3992 * stability of struct/union canonical relationship for equivalence checks).
3993 */
3994static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
3995{
3996        __u32 canon_type_id, targ_type_id;
3997        __u16 t_kind, c_kind;
3998        __u32 t_id, c_id;
3999        int i;
4000
4001        for (i = 0; i < d->hypot_cnt; i++) {
4002                canon_type_id = d->hypot_list[i];
4003                targ_type_id = d->hypot_map[canon_type_id];
4004                t_id = resolve_type_id(d, targ_type_id);
4005                c_id = resolve_type_id(d, canon_type_id);
4006                t_kind = btf_kind(btf__type_by_id(d->btf, t_id));
4007                c_kind = btf_kind(btf__type_by_id(d->btf, c_id));
4008                /*
4009                 * Resolve FWD into STRUCT/UNION.
4010                 * It's ok to resolve FWD into STRUCT/UNION that's not yet
4011                 * mapped to canonical representative (as opposed to
4012                 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
4013                 * eventually that struct is going to be mapped and all resolved
4014                 * FWDs will automatically resolve to correct canonical
4015                 * representative. This will happen before ref type deduping,
4016                 * which critically depends on stability of these mapping. This
4017                 * stability is not a requirement for STRUCT/UNION equivalence
4018                 * checks, though.
4019                 */
4020
4021                /* if it's the split BTF case, we still need to point base FWD
4022                 * to STRUCT/UNION in a split BTF, because FWDs from split BTF
4023                 * will be resolved against base FWD. If we don't point base
4024                 * canonical FWD to the resolved STRUCT/UNION, then all the
4025                 * FWDs in split BTF won't be correctly resolved to a proper
4026                 * STRUCT/UNION.
4027                 */
4028                if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
4029                        d->map[c_id] = t_id;
4030
4031                /* if graph equivalence determined that we'd need to adjust
4032                 * base canonical types, then we need to only point base FWDs
4033                 * to STRUCTs/UNIONs and do no more modifications. For all
4034                 * other purposes the type graphs were not equivalent.
4035                 */
4036                if (d->hypot_adjust_canon)
4037                        continue;
4038
4039                if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
4040                        d->map[t_id] = c_id;
4041
4042                if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
4043                    c_kind != BTF_KIND_FWD &&
4044                    is_type_mapped(d, c_id) &&
4045                    !is_type_mapped(d, t_id)) {
4046                        /*
4047                         * as a perf optimization, we can map struct/union
4048                         * that's part of type graph we just verified for
4049                         * equivalence. We can do that for struct/union that has
4050                         * canonical representative only, though.
4051                         */
4052                        d->map[t_id] = c_id;
4053                }
4054        }
4055}
4056
4057/*
4058 * Deduplicate struct/union types.
4059 *
4060 * For each struct/union type its type signature hash is calculated, taking
4061 * into account type's name, size, number, order and names of fields, but
4062 * ignoring type ID's referenced from fields, because they might not be deduped
4063 * completely until after reference types deduplication phase. This type hash
4064 * is used to iterate over all potential canonical types, sharing same hash.
4065 * For each canonical candidate we check whether type graphs that they form
4066 * (through referenced types in fields and so on) are equivalent using algorithm
4067 * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
4068 * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
4069 * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
4070 * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
4071 * potentially map other structs/unions to their canonical representatives,
4072 * if such relationship hasn't yet been established. This speeds up algorithm
4073 * by eliminating some of the duplicate work.
4074 *
4075 * If no matching canonical representative was found, struct/union is marked
4076 * as canonical for itself and is added into btf_dedup->dedup_table hash map
4077 * for further look ups.
4078 */
4079static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
4080{
4081        struct btf_type *cand_type, *t;
4082        struct hashmap_entry *hash_entry;
4083        /* if we don't find equivalent type, then we are canonical */
4084        __u32 new_id = type_id;
4085        __u16 kind;
4086        long h;
4087
4088        /* already deduped or is in process of deduping (loop detected) */
4089        if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4090                return 0;
4091
4092        t = btf_type_by_id(d->btf, type_id);
4093        kind = btf_kind(t);
4094
4095        if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4096                return 0;
4097
4098        h = btf_hash_struct(t);
4099        for_each_dedup_cand(d, hash_entry, h) {
4100                __u32 cand_id = (__u32)(long)hash_entry->value;
4101                int eq;
4102
4103                /*
4104                 * Even though btf_dedup_is_equiv() checks for
4105                 * btf_shallow_equal_struct() internally when checking two
4106                 * structs (unions) for equivalence, we need to guard here
4107                 * from picking matching FWD type as a dedup candidate.
4108                 * This can happen due to hash collision. In such case just
4109                 * relying on btf_dedup_is_equiv() would lead to potentially
4110                 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
4111                 * FWD and compatible STRUCT/UNION are considered equivalent.
4112                 */
4113                cand_type = btf_type_by_id(d->btf, cand_id);
4114                if (!btf_shallow_equal_struct(t, cand_type))
4115                        continue;
4116
4117                btf_dedup_clear_hypot_map(d);
4118                eq = btf_dedup_is_equiv(d, type_id, cand_id);
4119                if (eq < 0)
4120                        return eq;
4121                if (!eq)
4122                        continue;
4123                btf_dedup_merge_hypot_map(d);
4124                if (d->hypot_adjust_canon) /* not really equivalent */
4125                        continue;
4126                new_id = cand_id;
4127                break;
4128        }
4129
4130        d->map[type_id] = new_id;
4131        if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4132                return -ENOMEM;
4133
4134        return 0;
4135}
4136
4137static int btf_dedup_struct_types(struct btf_dedup *d)
4138{
4139        int i, err;
4140
4141        for (i = 0; i < d->btf->nr_types; i++) {
4142                err = btf_dedup_struct_type(d, d->btf->start_id + i);
4143                if (err)
4144                        return err;
4145        }
4146        return 0;
4147}
4148
4149/*
4150 * Deduplicate reference type.
4151 *
4152 * Once all primitive and struct/union types got deduplicated, we can easily
4153 * deduplicate all other (reference) BTF types. This is done in two steps:
4154 *
4155 * 1. Resolve all referenced type IDs into their canonical type IDs. This
4156 * resolution can be done either immediately for primitive or struct/union types
4157 * (because they were deduped in previous two phases) or recursively for
4158 * reference types. Recursion will always terminate at either primitive or
4159 * struct/union type, at which point we can "unwind" chain of reference types
4160 * one by one. There is no danger of encountering cycles because in C type
4161 * system the only way to form type cycle is through struct/union, so any chain
4162 * of reference types, even those taking part in a type cycle, will inevitably
4163 * reach struct/union at some point.
4164 *
4165 * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
4166 * becomes "stable", in the sense that no further deduplication will cause
4167 * any changes to it. With that, it's now possible to calculate type's signature
4168 * hash (this time taking into account referenced type IDs) and loop over all
4169 * potential canonical representatives. If no match was found, current type
4170 * will become canonical representative of itself and will be added into
4171 * btf_dedup->dedup_table as another possible canonical representative.
4172 */
4173static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
4174{
4175        struct hashmap_entry *hash_entry;
4176        __u32 new_id = type_id, cand_id;
4177        struct btf_type *t, *cand;
4178        /* if we don't find equivalent type, then we are representative type */
4179        int ref_type_id;
4180        long h;
4181
4182        if (d->map[type_id] == BTF_IN_PROGRESS_ID)
4183                return -ELOOP;
4184        if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4185                return resolve_type_id(d, type_id);
4186
4187        t = btf_type_by_id(d->btf, type_id);
4188        d->map[type_id] = BTF_IN_PROGRESS_ID;
4189
4190        switch (btf_kind(t)) {
4191        case BTF_KIND_CONST:
4192        case BTF_KIND_VOLATILE:
4193        case BTF_KIND_RESTRICT:
4194        case BTF_KIND_PTR:
4195        case BTF_KIND_TYPEDEF:
4196        case BTF_KIND_FUNC:
4197                ref_type_id = btf_dedup_ref_type(d, t->type);
4198                if (ref_type_id < 0)
4199                        return ref_type_id;
4200                t->type = ref_type_id;
4201
4202                h = btf_hash_common(t);
4203                for_each_dedup_cand(d, hash_entry, h) {
4204                        cand_id = (__u32)(long)hash_entry->value;
4205                        cand = btf_type_by_id(d->btf, cand_id);
4206                        if (btf_equal_common(t, cand)) {
4207                                new_id = cand_id;
4208                                break;
4209                        }
4210                }
4211                break;
4212
4213        case BTF_KIND_ARRAY: {
4214                struct btf_array *info = btf_array(t);
4215
4216                ref_type_id = btf_dedup_ref_type(d, info->type);
4217                if (ref_type_id < 0)
4218                        return ref_type_id;
4219                info->type = ref_type_id;
4220
4221                ref_type_id = btf_dedup_ref_type(d, info->index_type);
4222                if (ref_type_id < 0)
4223                        return ref_type_id;
4224                info->index_type = ref_type_id;
4225
4226                h = btf_hash_array(t);
4227                for_each_dedup_cand(d, hash_entry, h) {
4228                        cand_id = (__u32)(long)hash_entry->value;
4229                        cand = btf_type_by_id(d->btf, cand_id);
4230                        if (btf_equal_array(t, cand)) {
4231                                new_id = cand_id;
4232                                break;
4233                        }
4234                }
4235                break;
4236        }
4237
4238        case BTF_KIND_FUNC_PROTO: {
4239                struct btf_param *param;
4240                __u16 vlen;
4241                int i;
4242
4243                ref_type_id = btf_dedup_ref_type(d, t->type);
4244                if (ref_type_id < 0)
4245                        return ref_type_id;
4246                t->type = ref_type_id;
4247
4248                vlen = btf_vlen(t);
4249                param = btf_params(t);
4250                for (i = 0; i < vlen; i++) {
4251                        ref_type_id = btf_dedup_ref_type(d, param->type);
4252                        if (ref_type_id < 0)
4253                                return ref_type_id;
4254                        param->type = ref_type_id;
4255                        param++;
4256                }
4257
4258                h = btf_hash_fnproto(t);
4259                for_each_dedup_cand(d, hash_entry, h) {
4260                        cand_id = (__u32)(long)hash_entry->value;
4261                        cand = btf_type_by_id(d->btf, cand_id);
4262                        if (btf_equal_fnproto(t, cand)) {
4263                                new_id = cand_id;
4264                                break;
4265                        }
4266                }
4267                break;
4268        }
4269
4270        default:
4271                return -EINVAL;
4272        }
4273
4274        d->map[type_id] = new_id;
4275        if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4276                return -ENOMEM;
4277
4278        return new_id;
4279}
4280
4281static int btf_dedup_ref_types(struct btf_dedup *d)
4282{
4283        int i, err;
4284
4285        for (i = 0; i < d->btf->nr_types; i++) {
4286                err = btf_dedup_ref_type(d, d->btf->start_id + i);
4287                if (err < 0)
4288                        return err;
4289        }
4290        /* we won't need d->dedup_table anymore */
4291        hashmap__free(d->dedup_table);
4292        d->dedup_table = NULL;
4293        return 0;
4294}
4295
4296/*
4297 * Compact types.
4298 *
4299 * After we established for each type its corresponding canonical representative
4300 * type, we now can eliminate types that are not canonical and leave only
4301 * canonical ones layed out sequentially in memory by copying them over
4302 * duplicates. During compaction btf_dedup->hypot_map array is reused to store
4303 * a map from original type ID to a new compacted type ID, which will be used
4304 * during next phase to "fix up" type IDs, referenced from struct/union and
4305 * reference types.
4306 */
4307static int btf_dedup_compact_types(struct btf_dedup *d)
4308{
4309        __u32 *new_offs;
4310        __u32 next_type_id = d->btf->start_id;
4311        const struct btf_type *t;
4312        void *p;
4313        int i, id, len;
4314
4315        /* we are going to reuse hypot_map to store compaction remapping */
4316        d->hypot_map[0] = 0;
4317        /* base BTF types are not renumbered */
4318        for (id = 1; id < d->btf->start_id; id++)
4319                d->hypot_map[id] = id;
4320        for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++)
4321                d->hypot_map[id] = BTF_UNPROCESSED_ID;
4322
4323        p = d->btf->types_data;
4324
4325        for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++) {
4326                if (d->map[id] != id)
4327                        continue;
4328
4329                t = btf__type_by_id(d->btf, id);
4330                len = btf_type_size(t);
4331                if (len < 0)
4332                        return len;
4333
4334                memmove(p, t, len);
4335                d->hypot_map[id] = next_type_id;
4336                d->btf->type_offs[next_type_id - d->btf->start_id] = p - d->btf->types_data;
4337                p += len;
4338                next_type_id++;
4339        }
4340
4341        /* shrink struct btf's internal types index and update btf_header */
4342        d->btf->nr_types = next_type_id - d->btf->start_id;
4343        d->btf->type_offs_cap = d->btf->nr_types;
4344        d->btf->hdr->type_len = p - d->btf->types_data;
4345        new_offs = libbpf_reallocarray(d->btf->type_offs, d->btf->type_offs_cap,
4346                                       sizeof(*new_offs));
4347        if (d->btf->type_offs_cap && !new_offs)
4348                return -ENOMEM;
4349        d->btf->type_offs = new_offs;
4350        d->btf->hdr->str_off = d->btf->hdr->type_len;
4351        d->btf->raw_size = d->btf->hdr->hdr_len + d->btf->hdr->type_len + d->btf->hdr->str_len;
4352        return 0;
4353}
4354
4355/*
4356 * Figure out final (deduplicated and compacted) type ID for provided original
4357 * `type_id` by first resolving it into corresponding canonical type ID and
4358 * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
4359 * which is populated during compaction phase.
4360 */
4361static int btf_dedup_remap_type_id(__u32 *type_id, void *ctx)
4362{
4363        struct btf_dedup *d = ctx;
4364        __u32 resolved_type_id, new_type_id;
4365
4366        resolved_type_id = resolve_type_id(d, *type_id);
4367        new_type_id = d->hypot_map[resolved_type_id];
4368        if (new_type_id > BTF_MAX_NR_TYPES)
4369                return -EINVAL;
4370
4371        *type_id = new_type_id;
4372        return 0;
4373}
4374
4375/*
4376 * Remap referenced type IDs into deduped type IDs.
4377 *
4378 * After BTF types are deduplicated and compacted, their final type IDs may
4379 * differ from original ones. The map from original to a corresponding
4380 * deduped type ID is stored in btf_dedup->hypot_map and is populated during
4381 * compaction phase. During remapping phase we are rewriting all type IDs
4382 * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
4383 * their final deduped type IDs.
4384 */
4385static int btf_dedup_remap_types(struct btf_dedup *d)
4386{
4387        int i, r;
4388
4389        for (i = 0; i < d->btf->nr_types; i++) {
4390                struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
4391
4392                r = btf_type_visit_type_ids(t, btf_dedup_remap_type_id, d);
4393                if (r)
4394                        return r;
4395        }
4396
4397        if (!d->btf_ext)
4398                return 0;
4399
4400        r = btf_ext_visit_type_ids(d->btf_ext, btf_dedup_remap_type_id, d);
4401        if (r)
4402                return r;
4403
4404        return 0;
4405}
4406
4407/*
4408 * Probe few well-known locations for vmlinux kernel image and try to load BTF
4409 * data out of it to use for target BTF.
4410 */
4411struct btf *btf__load_vmlinux_btf(void)
4412{
4413        struct {
4414                const char *path_fmt;
4415                bool raw_btf;
4416        } locations[] = {
4417                /* try canonical vmlinux BTF through sysfs first */
4418                { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
4419                /* fall back to trying to find vmlinux ELF on disk otherwise */
4420                { "/boot/vmlinux-%1$s" },
4421                { "/lib/modules/%1$s/vmlinux-%1$s" },
4422                { "/lib/modules/%1$s/build/vmlinux" },
4423                { "/usr/lib/modules/%1$s/kernel/vmlinux" },
4424                { "/usr/lib/debug/boot/vmlinux-%1$s" },
4425                { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
4426                { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
4427        };
4428        char path[PATH_MAX + 1];
4429        struct utsname buf;
4430        struct btf *btf;
4431        int i, err;
4432
4433        uname(&buf);
4434
4435        for (i = 0; i < ARRAY_SIZE(locations); i++) {
4436                snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
4437
4438                if (access(path, R_OK))
4439                        continue;
4440
4441                if (locations[i].raw_btf)
4442                        btf = btf__parse_raw(path);
4443                else
4444                        btf = btf__parse_elf(path, NULL);
4445                err = libbpf_get_error(btf);
4446                pr_debug("loading kernel BTF '%s': %d\n", path, err);
4447                if (err)
4448                        continue;
4449
4450                return btf;
4451        }
4452
4453        pr_warn("failed to find valid kernel BTF\n");
4454        return libbpf_err_ptr(-ESRCH);
4455}
4456
4457struct btf *libbpf_find_kernel_btf(void) __attribute__((alias("btf__load_vmlinux_btf")));
4458
4459struct btf *btf__load_module_btf(const char *module_name, struct btf *vmlinux_btf)
4460{
4461        char path[80];
4462
4463        snprintf(path, sizeof(path), "/sys/kernel/btf/%s", module_name);
4464        return btf__parse_split(path, vmlinux_btf);
4465}
4466
4467int btf_type_visit_type_ids(struct btf_type *t, type_id_visit_fn visit, void *ctx)
4468{
4469        int i, n, err;
4470
4471        switch (btf_kind(t)) {
4472        case BTF_KIND_INT:
4473        case BTF_KIND_FLOAT:
4474        case BTF_KIND_ENUM:
4475                return 0;
4476
4477        case BTF_KIND_FWD:
4478        case BTF_KIND_CONST:
4479        case BTF_KIND_VOLATILE:
4480        case BTF_KIND_RESTRICT:
4481        case BTF_KIND_PTR:
4482        case BTF_KIND_TYPEDEF:
4483        case BTF_KIND_FUNC:
4484        case BTF_KIND_VAR:
4485                return visit(&t->type, ctx);
4486
4487        case BTF_KIND_ARRAY: {
4488                struct btf_array *a = btf_array(t);
4489
4490                err = visit(&a->type, ctx);
4491                err = err ?: visit(&a->index_type, ctx);
4492                return err;
4493        }
4494
4495        case BTF_KIND_STRUCT:
4496        case BTF_KIND_UNION: {
4497                struct btf_member *m = btf_members(t);
4498
4499                for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4500                        err = visit(&m->type, ctx);
4501                        if (err)
4502                                return err;
4503                }
4504                return 0;
4505        }
4506
4507        case BTF_KIND_FUNC_PROTO: {
4508                struct btf_param *m = btf_params(t);
4509
4510                err = visit(&t->type, ctx);
4511                if (err)
4512                        return err;
4513                for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4514                        err = visit(&m->type, ctx);
4515                        if (err)
4516                                return err;
4517                }
4518                return 0;
4519        }
4520
4521        case BTF_KIND_DATASEC: {
4522                struct btf_var_secinfo *m = btf_var_secinfos(t);
4523
4524                for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4525                        err = visit(&m->type, ctx);
4526                        if (err)
4527                                return err;
4528                }
4529                return 0;
4530        }
4531
4532        default:
4533                return -EINVAL;
4534        }
4535}
4536
4537int btf_type_visit_str_offs(struct btf_type *t, str_off_visit_fn visit, void *ctx)
4538{
4539        int i, n, err;
4540
4541        err = visit(&t->name_off, ctx);
4542        if (err)
4543                return err;
4544
4545        switch (btf_kind(t)) {
4546        case BTF_KIND_STRUCT:
4547        case BTF_KIND_UNION: {
4548                struct btf_member *m = btf_members(t);
4549
4550                for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4551                        err = visit(&m->name_off, ctx);
4552                        if (err)
4553                                return err;
4554                }
4555                break;
4556        }
4557        case BTF_KIND_ENUM: {
4558                struct btf_enum *m = btf_enum(t);
4559
4560                for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4561                        err = visit(&m->name_off, ctx);
4562                        if (err)
4563                                return err;
4564                }
4565                break;
4566        }
4567        case BTF_KIND_FUNC_PROTO: {
4568                struct btf_param *m = btf_params(t);
4569
4570                for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4571                        err = visit(&m->name_off, ctx);
4572                        if (err)
4573                                return err;
4574                }
4575                break;
4576        }
4577        default:
4578                break;
4579        }
4580
4581        return 0;
4582}
4583
4584int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx)
4585{
4586        const struct btf_ext_info *seg;
4587        struct btf_ext_info_sec *sec;
4588        int i, err;
4589
4590        seg = &btf_ext->func_info;
4591        for_each_btf_ext_sec(seg, sec) {
4592                struct bpf_func_info_min *rec;
4593
4594                for_each_btf_ext_rec(seg, sec, i, rec) {
4595                        err = visit(&rec->type_id, ctx);
4596                        if (err < 0)
4597                                return err;
4598                }
4599        }
4600
4601        seg = &btf_ext->core_relo_info;
4602        for_each_btf_ext_sec(seg, sec) {
4603                struct bpf_core_relo *rec;
4604
4605                for_each_btf_ext_rec(seg, sec, i, rec) {
4606                        err = visit(&rec->type_id, ctx);
4607                        if (err < 0)
4608                                return err;
4609                }
4610        }
4611
4612        return 0;
4613}
4614
4615int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx)
4616{
4617        const struct btf_ext_info *seg;
4618        struct btf_ext_info_sec *sec;
4619        int i, err;
4620
4621        seg = &btf_ext->func_info;
4622        for_each_btf_ext_sec(seg, sec) {
4623                err = visit(&sec->sec_name_off, ctx);
4624                if (err)
4625                        return err;
4626        }
4627
4628        seg = &btf_ext->line_info;
4629        for_each_btf_ext_sec(seg, sec) {
4630                struct bpf_line_info_min *rec;
4631
4632                err = visit(&sec->sec_name_off, ctx);
4633                if (err)
4634                        return err;
4635
4636                for_each_btf_ext_rec(seg, sec, i, rec) {
4637                        err = visit(&rec->file_name_off, ctx);
4638                        if (err)
4639                                return err;
4640                        err = visit(&rec->line_off, ctx);
4641                        if (err)
4642                                return err;
4643                }
4644        }
4645
4646        seg = &btf_ext->core_relo_info;
4647        for_each_btf_ext_sec(seg, sec) {
4648                struct bpf_core_relo *rec;
4649
4650                err = visit(&sec->sec_name_off, ctx);
4651                if (err)
4652                        return err;
4653
4654                for_each_btf_ext_rec(seg, sec, i, rec) {
4655                        err = visit(&rec->access_str_off, ctx);
4656                        if (err)
4657                                return err;
4658                }
4659        }
4660
4661        return 0;
4662}
4663