xref: /openbmc/linux/tools/lib/bpf/btf.c (revision 1c4b5ecb)
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 
29 static struct btf_type btf_void;
30 
31 struct 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__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 
126 static 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  */
141 void *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  */
178 int 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 
192 static void *btf_add_type_offs_mem(struct btf *btf, size_t add_cnt)
193 {
194 	return libbpf_add_mem((void **)&btf->type_offs, &btf->type_offs_cap, sizeof(__u32),
195 			      btf->nr_types, BTF_MAX_NR_TYPES, add_cnt);
196 }
197 
198 static int btf_add_type_idx_entry(struct btf *btf, __u32 type_off)
199 {
200 	__u32 *p;
201 
202 	p = btf_add_type_offs_mem(btf, 1);
203 	if (!p)
204 		return -ENOMEM;
205 
206 	*p = type_off;
207 	return 0;
208 }
209 
210 static void btf_bswap_hdr(struct btf_header *h)
211 {
212 	h->magic = bswap_16(h->magic);
213 	h->hdr_len = bswap_32(h->hdr_len);
214 	h->type_off = bswap_32(h->type_off);
215 	h->type_len = bswap_32(h->type_len);
216 	h->str_off = bswap_32(h->str_off);
217 	h->str_len = bswap_32(h->str_len);
218 }
219 
220 static int btf_parse_hdr(struct btf *btf)
221 {
222 	struct btf_header *hdr = btf->hdr;
223 	__u32 meta_left;
224 
225 	if (btf->raw_size < sizeof(struct btf_header)) {
226 		pr_debug("BTF header not found\n");
227 		return -EINVAL;
228 	}
229 
230 	if (hdr->magic == bswap_16(BTF_MAGIC)) {
231 		btf->swapped_endian = true;
232 		if (bswap_32(hdr->hdr_len) != sizeof(struct btf_header)) {
233 			pr_warn("Can't load BTF with non-native endianness due to unsupported header length %u\n",
234 				bswap_32(hdr->hdr_len));
235 			return -ENOTSUP;
236 		}
237 		btf_bswap_hdr(hdr);
238 	} else if (hdr->magic != BTF_MAGIC) {
239 		pr_debug("Invalid BTF magic: %x\n", hdr->magic);
240 		return -EINVAL;
241 	}
242 
243 	if (btf->raw_size < hdr->hdr_len) {
244 		pr_debug("BTF header len %u larger than data size %u\n",
245 			 hdr->hdr_len, btf->raw_size);
246 		return -EINVAL;
247 	}
248 
249 	meta_left = btf->raw_size - hdr->hdr_len;
250 	if (meta_left < (long long)hdr->str_off + hdr->str_len) {
251 		pr_debug("Invalid BTF total size: %u\n", btf->raw_size);
252 		return -EINVAL;
253 	}
254 
255 	if ((long long)hdr->type_off + hdr->type_len > hdr->str_off) {
256 		pr_debug("Invalid BTF data sections layout: type data at %u + %u, strings data at %u + %u\n",
257 			 hdr->type_off, hdr->type_len, hdr->str_off, hdr->str_len);
258 		return -EINVAL;
259 	}
260 
261 	if (hdr->type_off % 4) {
262 		pr_debug("BTF type section is not aligned to 4 bytes\n");
263 		return -EINVAL;
264 	}
265 
266 	return 0;
267 }
268 
269 static int btf_parse_str_sec(struct btf *btf)
270 {
271 	const struct btf_header *hdr = btf->hdr;
272 	const char *start = btf->strs_data;
273 	const char *end = start + btf->hdr->str_len;
274 
275 	if (btf->base_btf && hdr->str_len == 0)
276 		return 0;
277 	if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET || end[-1]) {
278 		pr_debug("Invalid BTF string section\n");
279 		return -EINVAL;
280 	}
281 	if (!btf->base_btf && start[0]) {
282 		pr_debug("Invalid BTF string section\n");
283 		return -EINVAL;
284 	}
285 	return 0;
286 }
287 
288 static int btf_type_size(const struct btf_type *t)
289 {
290 	const int base_size = sizeof(struct btf_type);
291 	__u16 vlen = btf_vlen(t);
292 
293 	switch (btf_kind(t)) {
294 	case BTF_KIND_FWD:
295 	case BTF_KIND_CONST:
296 	case BTF_KIND_VOLATILE:
297 	case BTF_KIND_RESTRICT:
298 	case BTF_KIND_PTR:
299 	case BTF_KIND_TYPEDEF:
300 	case BTF_KIND_FUNC:
301 	case BTF_KIND_FLOAT:
302 	case BTF_KIND_TYPE_TAG:
303 		return base_size;
304 	case BTF_KIND_INT:
305 		return base_size + sizeof(__u32);
306 	case BTF_KIND_ENUM:
307 		return base_size + vlen * sizeof(struct btf_enum);
308 	case BTF_KIND_ARRAY:
309 		return base_size + sizeof(struct btf_array);
310 	case BTF_KIND_STRUCT:
311 	case BTF_KIND_UNION:
312 		return base_size + vlen * sizeof(struct btf_member);
313 	case BTF_KIND_FUNC_PROTO:
314 		return base_size + vlen * sizeof(struct btf_param);
315 	case BTF_KIND_VAR:
316 		return base_size + sizeof(struct btf_var);
317 	case BTF_KIND_DATASEC:
318 		return base_size + vlen * sizeof(struct btf_var_secinfo);
319 	case BTF_KIND_DECL_TAG:
320 		return base_size + sizeof(struct btf_decl_tag);
321 	default:
322 		pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
323 		return -EINVAL;
324 	}
325 }
326 
327 static void btf_bswap_type_base(struct btf_type *t)
328 {
329 	t->name_off = bswap_32(t->name_off);
330 	t->info = bswap_32(t->info);
331 	t->type = bswap_32(t->type);
332 }
333 
334 static int btf_bswap_type_rest(struct btf_type *t)
335 {
336 	struct btf_var_secinfo *v;
337 	struct btf_member *m;
338 	struct btf_array *a;
339 	struct btf_param *p;
340 	struct btf_enum *e;
341 	__u16 vlen = btf_vlen(t);
342 	int i;
343 
344 	switch (btf_kind(t)) {
345 	case BTF_KIND_FWD:
346 	case BTF_KIND_CONST:
347 	case BTF_KIND_VOLATILE:
348 	case BTF_KIND_RESTRICT:
349 	case BTF_KIND_PTR:
350 	case BTF_KIND_TYPEDEF:
351 	case BTF_KIND_FUNC:
352 	case BTF_KIND_FLOAT:
353 	case BTF_KIND_TYPE_TAG:
354 		return 0;
355 	case BTF_KIND_INT:
356 		*(__u32 *)(t + 1) = bswap_32(*(__u32 *)(t + 1));
357 		return 0;
358 	case BTF_KIND_ENUM:
359 		for (i = 0, e = btf_enum(t); i < vlen; i++, e++) {
360 			e->name_off = bswap_32(e->name_off);
361 			e->val = bswap_32(e->val);
362 		}
363 		return 0;
364 	case BTF_KIND_ARRAY:
365 		a = btf_array(t);
366 		a->type = bswap_32(a->type);
367 		a->index_type = bswap_32(a->index_type);
368 		a->nelems = bswap_32(a->nelems);
369 		return 0;
370 	case BTF_KIND_STRUCT:
371 	case BTF_KIND_UNION:
372 		for (i = 0, m = btf_members(t); i < vlen; i++, m++) {
373 			m->name_off = bswap_32(m->name_off);
374 			m->type = bswap_32(m->type);
375 			m->offset = bswap_32(m->offset);
376 		}
377 		return 0;
378 	case BTF_KIND_FUNC_PROTO:
379 		for (i = 0, p = btf_params(t); i < vlen; i++, p++) {
380 			p->name_off = bswap_32(p->name_off);
381 			p->type = bswap_32(p->type);
382 		}
383 		return 0;
384 	case BTF_KIND_VAR:
385 		btf_var(t)->linkage = bswap_32(btf_var(t)->linkage);
386 		return 0;
387 	case BTF_KIND_DATASEC:
388 		for (i = 0, v = btf_var_secinfos(t); i < vlen; i++, v++) {
389 			v->type = bswap_32(v->type);
390 			v->offset = bswap_32(v->offset);
391 			v->size = bswap_32(v->size);
392 		}
393 		return 0;
394 	case BTF_KIND_DECL_TAG:
395 		btf_decl_tag(t)->component_idx = bswap_32(btf_decl_tag(t)->component_idx);
396 		return 0;
397 	default:
398 		pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
399 		return -EINVAL;
400 	}
401 }
402 
403 static int btf_parse_type_sec(struct btf *btf)
404 {
405 	struct btf_header *hdr = btf->hdr;
406 	void *next_type = btf->types_data;
407 	void *end_type = next_type + hdr->type_len;
408 	int err, type_size;
409 
410 	while (next_type + sizeof(struct btf_type) <= end_type) {
411 		if (btf->swapped_endian)
412 			btf_bswap_type_base(next_type);
413 
414 		type_size = btf_type_size(next_type);
415 		if (type_size < 0)
416 			return type_size;
417 		if (next_type + type_size > end_type) {
418 			pr_warn("BTF type [%d] is malformed\n", btf->start_id + btf->nr_types);
419 			return -EINVAL;
420 		}
421 
422 		if (btf->swapped_endian && btf_bswap_type_rest(next_type))
423 			return -EINVAL;
424 
425 		err = btf_add_type_idx_entry(btf, next_type - btf->types_data);
426 		if (err)
427 			return err;
428 
429 		next_type += type_size;
430 		btf->nr_types++;
431 	}
432 
433 	if (next_type != end_type) {
434 		pr_warn("BTF types data is malformed\n");
435 		return -EINVAL;
436 	}
437 
438 	return 0;
439 }
440 
441 __u32 btf__get_nr_types(const struct btf *btf)
442 {
443 	return btf->start_id + btf->nr_types - 1;
444 }
445 
446 __u32 btf__type_cnt(const struct btf *btf)
447 {
448 	return btf->start_id + btf->nr_types;
449 }
450 
451 const struct btf *btf__base_btf(const struct btf *btf)
452 {
453 	return btf->base_btf;
454 }
455 
456 /* internal helper returning non-const pointer to a type */
457 struct btf_type *btf_type_by_id(const struct btf *btf, __u32 type_id)
458 {
459 	if (type_id == 0)
460 		return &btf_void;
461 	if (type_id < btf->start_id)
462 		return btf_type_by_id(btf->base_btf, type_id);
463 	return btf->types_data + btf->type_offs[type_id - btf->start_id];
464 }
465 
466 const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
467 {
468 	if (type_id >= btf->start_id + btf->nr_types)
469 		return errno = EINVAL, NULL;
470 	return btf_type_by_id((struct btf *)btf, type_id);
471 }
472 
473 static int determine_ptr_size(const struct btf *btf)
474 {
475 	const struct btf_type *t;
476 	const char *name;
477 	int i, n;
478 
479 	if (btf->base_btf && btf->base_btf->ptr_sz > 0)
480 		return btf->base_btf->ptr_sz;
481 
482 	n = btf__type_cnt(btf);
483 	for (i = 1; i < n; i++) {
484 		t = btf__type_by_id(btf, i);
485 		if (!btf_is_int(t))
486 			continue;
487 
488 		name = btf__name_by_offset(btf, t->name_off);
489 		if (!name)
490 			continue;
491 
492 		if (strcmp(name, "long int") == 0 ||
493 		    strcmp(name, "long unsigned int") == 0) {
494 			if (t->size != 4 && t->size != 8)
495 				continue;
496 			return t->size;
497 		}
498 	}
499 
500 	return -1;
501 }
502 
503 static size_t btf_ptr_sz(const struct btf *btf)
504 {
505 	if (!btf->ptr_sz)
506 		((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
507 	return btf->ptr_sz < 0 ? sizeof(void *) : btf->ptr_sz;
508 }
509 
510 /* Return pointer size this BTF instance assumes. The size is heuristically
511  * determined by looking for 'long' or 'unsigned long' integer type and
512  * recording its size in bytes. If BTF type information doesn't have any such
513  * type, this function returns 0. In the latter case, native architecture's
514  * pointer size is assumed, so will be either 4 or 8, depending on
515  * architecture that libbpf was compiled for. It's possible to override
516  * guessed value by using btf__set_pointer_size() API.
517  */
518 size_t btf__pointer_size(const struct btf *btf)
519 {
520 	if (!btf->ptr_sz)
521 		((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
522 
523 	if (btf->ptr_sz < 0)
524 		/* not enough BTF type info to guess */
525 		return 0;
526 
527 	return btf->ptr_sz;
528 }
529 
530 /* Override or set pointer size in bytes. Only values of 4 and 8 are
531  * supported.
532  */
533 int btf__set_pointer_size(struct btf *btf, size_t ptr_sz)
534 {
535 	if (ptr_sz != 4 && ptr_sz != 8)
536 		return libbpf_err(-EINVAL);
537 	btf->ptr_sz = ptr_sz;
538 	return 0;
539 }
540 
541 static bool is_host_big_endian(void)
542 {
543 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
544 	return false;
545 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
546 	return true;
547 #else
548 # error "Unrecognized __BYTE_ORDER__"
549 #endif
550 }
551 
552 enum btf_endianness btf__endianness(const struct btf *btf)
553 {
554 	if (is_host_big_endian())
555 		return btf->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
556 	else
557 		return btf->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
558 }
559 
560 int btf__set_endianness(struct btf *btf, enum btf_endianness endian)
561 {
562 	if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
563 		return libbpf_err(-EINVAL);
564 
565 	btf->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
566 	if (!btf->swapped_endian) {
567 		free(btf->raw_data_swapped);
568 		btf->raw_data_swapped = NULL;
569 	}
570 	return 0;
571 }
572 
573 static bool btf_type_is_void(const struct btf_type *t)
574 {
575 	return t == &btf_void || btf_is_fwd(t);
576 }
577 
578 static bool btf_type_is_void_or_null(const struct btf_type *t)
579 {
580 	return !t || btf_type_is_void(t);
581 }
582 
583 #define MAX_RESOLVE_DEPTH 32
584 
585 __s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
586 {
587 	const struct btf_array *array;
588 	const struct btf_type *t;
589 	__u32 nelems = 1;
590 	__s64 size = -1;
591 	int i;
592 
593 	t = btf__type_by_id(btf, type_id);
594 	for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t); i++) {
595 		switch (btf_kind(t)) {
596 		case BTF_KIND_INT:
597 		case BTF_KIND_STRUCT:
598 		case BTF_KIND_UNION:
599 		case BTF_KIND_ENUM:
600 		case BTF_KIND_DATASEC:
601 		case BTF_KIND_FLOAT:
602 			size = t->size;
603 			goto done;
604 		case BTF_KIND_PTR:
605 			size = btf_ptr_sz(btf);
606 			goto done;
607 		case BTF_KIND_TYPEDEF:
608 		case BTF_KIND_VOLATILE:
609 		case BTF_KIND_CONST:
610 		case BTF_KIND_RESTRICT:
611 		case BTF_KIND_VAR:
612 		case BTF_KIND_DECL_TAG:
613 		case BTF_KIND_TYPE_TAG:
614 			type_id = t->type;
615 			break;
616 		case BTF_KIND_ARRAY:
617 			array = btf_array(t);
618 			if (nelems && array->nelems > UINT32_MAX / nelems)
619 				return libbpf_err(-E2BIG);
620 			nelems *= array->nelems;
621 			type_id = array->type;
622 			break;
623 		default:
624 			return libbpf_err(-EINVAL);
625 		}
626 
627 		t = btf__type_by_id(btf, type_id);
628 	}
629 
630 done:
631 	if (size < 0)
632 		return libbpf_err(-EINVAL);
633 	if (nelems && size > UINT32_MAX / nelems)
634 		return libbpf_err(-E2BIG);
635 
636 	return nelems * size;
637 }
638 
639 int btf__align_of(const struct btf *btf, __u32 id)
640 {
641 	const struct btf_type *t = btf__type_by_id(btf, id);
642 	__u16 kind = btf_kind(t);
643 
644 	switch (kind) {
645 	case BTF_KIND_INT:
646 	case BTF_KIND_ENUM:
647 	case BTF_KIND_FLOAT:
648 		return min(btf_ptr_sz(btf), (size_t)t->size);
649 	case BTF_KIND_PTR:
650 		return btf_ptr_sz(btf);
651 	case BTF_KIND_TYPEDEF:
652 	case BTF_KIND_VOLATILE:
653 	case BTF_KIND_CONST:
654 	case BTF_KIND_RESTRICT:
655 	case BTF_KIND_TYPE_TAG:
656 		return btf__align_of(btf, t->type);
657 	case BTF_KIND_ARRAY:
658 		return btf__align_of(btf, btf_array(t)->type);
659 	case BTF_KIND_STRUCT:
660 	case BTF_KIND_UNION: {
661 		const struct btf_member *m = btf_members(t);
662 		__u16 vlen = btf_vlen(t);
663 		int i, max_align = 1, align;
664 
665 		for (i = 0; i < vlen; i++, m++) {
666 			align = btf__align_of(btf, m->type);
667 			if (align <= 0)
668 				return libbpf_err(align);
669 			max_align = max(max_align, align);
670 		}
671 
672 		return max_align;
673 	}
674 	default:
675 		pr_warn("unsupported BTF_KIND:%u\n", btf_kind(t));
676 		return errno = EINVAL, 0;
677 	}
678 }
679 
680 int btf__resolve_type(const struct btf *btf, __u32 type_id)
681 {
682 	const struct btf_type *t;
683 	int depth = 0;
684 
685 	t = btf__type_by_id(btf, type_id);
686 	while (depth < MAX_RESOLVE_DEPTH &&
687 	       !btf_type_is_void_or_null(t) &&
688 	       (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
689 		type_id = t->type;
690 		t = btf__type_by_id(btf, type_id);
691 		depth++;
692 	}
693 
694 	if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
695 		return libbpf_err(-EINVAL);
696 
697 	return type_id;
698 }
699 
700 __s32 btf__find_by_name(const struct btf *btf, const char *type_name)
701 {
702 	__u32 i, nr_types = btf__type_cnt(btf);
703 
704 	if (!strcmp(type_name, "void"))
705 		return 0;
706 
707 	for (i = 1; i < nr_types; i++) {
708 		const struct btf_type *t = btf__type_by_id(btf, i);
709 		const char *name = btf__name_by_offset(btf, t->name_off);
710 
711 		if (name && !strcmp(type_name, name))
712 			return i;
713 	}
714 
715 	return libbpf_err(-ENOENT);
716 }
717 
718 static __s32 btf_find_by_name_kind(const struct btf *btf, int start_id,
719 				   const char *type_name, __u32 kind)
720 {
721 	__u32 i, nr_types = btf__type_cnt(btf);
722 
723 	if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
724 		return 0;
725 
726 	for (i = start_id; i < nr_types; i++) {
727 		const struct btf_type *t = btf__type_by_id(btf, i);
728 		const char *name;
729 
730 		if (btf_kind(t) != kind)
731 			continue;
732 		name = btf__name_by_offset(btf, t->name_off);
733 		if (name && !strcmp(type_name, name))
734 			return i;
735 	}
736 
737 	return libbpf_err(-ENOENT);
738 }
739 
740 __s32 btf__find_by_name_kind_own(const struct btf *btf, const char *type_name,
741 				 __u32 kind)
742 {
743 	return btf_find_by_name_kind(btf, btf->start_id, type_name, kind);
744 }
745 
746 __s32 btf__find_by_name_kind(const struct btf *btf, const char *type_name,
747 			     __u32 kind)
748 {
749 	return btf_find_by_name_kind(btf, 1, type_name, kind);
750 }
751 
752 static bool btf_is_modifiable(const struct btf *btf)
753 {
754 	return (void *)btf->hdr != btf->raw_data;
755 }
756 
757 void btf__free(struct btf *btf)
758 {
759 	if (IS_ERR_OR_NULL(btf))
760 		return;
761 
762 	if (btf->fd >= 0)
763 		close(btf->fd);
764 
765 	if (btf_is_modifiable(btf)) {
766 		/* if BTF was modified after loading, it will have a split
767 		 * in-memory representation for header, types, and strings
768 		 * sections, so we need to free all of them individually. It
769 		 * might still have a cached contiguous raw data present,
770 		 * which will be unconditionally freed below.
771 		 */
772 		free(btf->hdr);
773 		free(btf->types_data);
774 		strset__free(btf->strs_set);
775 	}
776 	free(btf->raw_data);
777 	free(btf->raw_data_swapped);
778 	free(btf->type_offs);
779 	free(btf);
780 }
781 
782 static struct btf *btf_new_empty(struct btf *base_btf)
783 {
784 	struct btf *btf;
785 
786 	btf = calloc(1, sizeof(*btf));
787 	if (!btf)
788 		return ERR_PTR(-ENOMEM);
789 
790 	btf->nr_types = 0;
791 	btf->start_id = 1;
792 	btf->start_str_off = 0;
793 	btf->fd = -1;
794 	btf->ptr_sz = sizeof(void *);
795 	btf->swapped_endian = false;
796 
797 	if (base_btf) {
798 		btf->base_btf = base_btf;
799 		btf->start_id = btf__type_cnt(base_btf);
800 		btf->start_str_off = base_btf->hdr->str_len;
801 	}
802 
803 	/* +1 for empty string at offset 0 */
804 	btf->raw_size = sizeof(struct btf_header) + (base_btf ? 0 : 1);
805 	btf->raw_data = calloc(1, btf->raw_size);
806 	if (!btf->raw_data) {
807 		free(btf);
808 		return ERR_PTR(-ENOMEM);
809 	}
810 
811 	btf->hdr = btf->raw_data;
812 	btf->hdr->hdr_len = sizeof(struct btf_header);
813 	btf->hdr->magic = BTF_MAGIC;
814 	btf->hdr->version = BTF_VERSION;
815 
816 	btf->types_data = btf->raw_data + btf->hdr->hdr_len;
817 	btf->strs_data = btf->raw_data + btf->hdr->hdr_len;
818 	btf->hdr->str_len = base_btf ? 0 : 1; /* empty string at offset 0 */
819 
820 	return btf;
821 }
822 
823 struct btf *btf__new_empty(void)
824 {
825 	return libbpf_ptr(btf_new_empty(NULL));
826 }
827 
828 struct btf *btf__new_empty_split(struct btf *base_btf)
829 {
830 	return libbpf_ptr(btf_new_empty(base_btf));
831 }
832 
833 static struct btf *btf_new(const void *data, __u32 size, struct btf *base_btf)
834 {
835 	struct btf *btf;
836 	int err;
837 
838 	btf = calloc(1, sizeof(struct btf));
839 	if (!btf)
840 		return ERR_PTR(-ENOMEM);
841 
842 	btf->nr_types = 0;
843 	btf->start_id = 1;
844 	btf->start_str_off = 0;
845 	btf->fd = -1;
846 
847 	if (base_btf) {
848 		btf->base_btf = base_btf;
849 		btf->start_id = btf__type_cnt(base_btf);
850 		btf->start_str_off = base_btf->hdr->str_len;
851 	}
852 
853 	btf->raw_data = malloc(size);
854 	if (!btf->raw_data) {
855 		err = -ENOMEM;
856 		goto done;
857 	}
858 	memcpy(btf->raw_data, data, size);
859 	btf->raw_size = size;
860 
861 	btf->hdr = btf->raw_data;
862 	err = btf_parse_hdr(btf);
863 	if (err)
864 		goto done;
865 
866 	btf->strs_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->str_off;
867 	btf->types_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->type_off;
868 
869 	err = btf_parse_str_sec(btf);
870 	err = err ?: btf_parse_type_sec(btf);
871 	if (err)
872 		goto done;
873 
874 done:
875 	if (err) {
876 		btf__free(btf);
877 		return ERR_PTR(err);
878 	}
879 
880 	return btf;
881 }
882 
883 struct btf *btf__new(const void *data, __u32 size)
884 {
885 	return libbpf_ptr(btf_new(data, size, NULL));
886 }
887 
888 static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,
889 				 struct btf_ext **btf_ext)
890 {
891 	Elf_Data *btf_data = NULL, *btf_ext_data = NULL;
892 	int err = 0, fd = -1, idx = 0;
893 	struct btf *btf = NULL;
894 	Elf_Scn *scn = NULL;
895 	Elf *elf = NULL;
896 	GElf_Ehdr ehdr;
897 	size_t shstrndx;
898 
899 	if (elf_version(EV_CURRENT) == EV_NONE) {
900 		pr_warn("failed to init libelf for %s\n", path);
901 		return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
902 	}
903 
904 	fd = open(path, O_RDONLY | O_CLOEXEC);
905 	if (fd < 0) {
906 		err = -errno;
907 		pr_warn("failed to open %s: %s\n", path, strerror(errno));
908 		return ERR_PTR(err);
909 	}
910 
911 	err = -LIBBPF_ERRNO__FORMAT;
912 
913 	elf = elf_begin(fd, ELF_C_READ, NULL);
914 	if (!elf) {
915 		pr_warn("failed to open %s as ELF file\n", path);
916 		goto done;
917 	}
918 	if (!gelf_getehdr(elf, &ehdr)) {
919 		pr_warn("failed to get EHDR from %s\n", path);
920 		goto done;
921 	}
922 
923 	if (elf_getshdrstrndx(elf, &shstrndx)) {
924 		pr_warn("failed to get section names section index for %s\n",
925 			path);
926 		goto done;
927 	}
928 
929 	if (!elf_rawdata(elf_getscn(elf, shstrndx), NULL)) {
930 		pr_warn("failed to get e_shstrndx from %s\n", path);
931 		goto done;
932 	}
933 
934 	while ((scn = elf_nextscn(elf, scn)) != NULL) {
935 		GElf_Shdr sh;
936 		char *name;
937 
938 		idx++;
939 		if (gelf_getshdr(scn, &sh) != &sh) {
940 			pr_warn("failed to get section(%d) header from %s\n",
941 				idx, path);
942 			goto done;
943 		}
944 		name = elf_strptr(elf, shstrndx, sh.sh_name);
945 		if (!name) {
946 			pr_warn("failed to get section(%d) name from %s\n",
947 				idx, path);
948 			goto done;
949 		}
950 		if (strcmp(name, BTF_ELF_SEC) == 0) {
951 			btf_data = elf_getdata(scn, 0);
952 			if (!btf_data) {
953 				pr_warn("failed to get section(%d, %s) data from %s\n",
954 					idx, name, path);
955 				goto done;
956 			}
957 			continue;
958 		} else if (btf_ext && strcmp(name, BTF_EXT_ELF_SEC) == 0) {
959 			btf_ext_data = elf_getdata(scn, 0);
960 			if (!btf_ext_data) {
961 				pr_warn("failed to get section(%d, %s) data from %s\n",
962 					idx, name, path);
963 				goto done;
964 			}
965 			continue;
966 		}
967 	}
968 
969 	err = 0;
970 
971 	if (!btf_data) {
972 		err = -ENOENT;
973 		goto done;
974 	}
975 	btf = btf_new(btf_data->d_buf, btf_data->d_size, base_btf);
976 	err = libbpf_get_error(btf);
977 	if (err)
978 		goto done;
979 
980 	switch (gelf_getclass(elf)) {
981 	case ELFCLASS32:
982 		btf__set_pointer_size(btf, 4);
983 		break;
984 	case ELFCLASS64:
985 		btf__set_pointer_size(btf, 8);
986 		break;
987 	default:
988 		pr_warn("failed to get ELF class (bitness) for %s\n", path);
989 		break;
990 	}
991 
992 	if (btf_ext && btf_ext_data) {
993 		*btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
994 		err = libbpf_get_error(*btf_ext);
995 		if (err)
996 			goto done;
997 	} else if (btf_ext) {
998 		*btf_ext = NULL;
999 	}
1000 done:
1001 	if (elf)
1002 		elf_end(elf);
1003 	close(fd);
1004 
1005 	if (!err)
1006 		return btf;
1007 
1008 	if (btf_ext)
1009 		btf_ext__free(*btf_ext);
1010 	btf__free(btf);
1011 
1012 	return ERR_PTR(err);
1013 }
1014 
1015 struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
1016 {
1017 	return libbpf_ptr(btf_parse_elf(path, NULL, btf_ext));
1018 }
1019 
1020 struct btf *btf__parse_elf_split(const char *path, struct btf *base_btf)
1021 {
1022 	return libbpf_ptr(btf_parse_elf(path, base_btf, NULL));
1023 }
1024 
1025 static struct btf *btf_parse_raw(const char *path, struct btf *base_btf)
1026 {
1027 	struct btf *btf = NULL;
1028 	void *data = NULL;
1029 	FILE *f = NULL;
1030 	__u16 magic;
1031 	int err = 0;
1032 	long sz;
1033 
1034 	f = fopen(path, "rb");
1035 	if (!f) {
1036 		err = -errno;
1037 		goto err_out;
1038 	}
1039 
1040 	/* check BTF magic */
1041 	if (fread(&magic, 1, sizeof(magic), f) < sizeof(magic)) {
1042 		err = -EIO;
1043 		goto err_out;
1044 	}
1045 	if (magic != BTF_MAGIC && magic != bswap_16(BTF_MAGIC)) {
1046 		/* definitely not a raw BTF */
1047 		err = -EPROTO;
1048 		goto err_out;
1049 	}
1050 
1051 	/* get file size */
1052 	if (fseek(f, 0, SEEK_END)) {
1053 		err = -errno;
1054 		goto err_out;
1055 	}
1056 	sz = ftell(f);
1057 	if (sz < 0) {
1058 		err = -errno;
1059 		goto err_out;
1060 	}
1061 	/* rewind to the start */
1062 	if (fseek(f, 0, SEEK_SET)) {
1063 		err = -errno;
1064 		goto err_out;
1065 	}
1066 
1067 	/* pre-alloc memory and read all of BTF data */
1068 	data = malloc(sz);
1069 	if (!data) {
1070 		err = -ENOMEM;
1071 		goto err_out;
1072 	}
1073 	if (fread(data, 1, sz, f) < sz) {
1074 		err = -EIO;
1075 		goto err_out;
1076 	}
1077 
1078 	/* finally parse BTF data */
1079 	btf = btf_new(data, sz, base_btf);
1080 
1081 err_out:
1082 	free(data);
1083 	if (f)
1084 		fclose(f);
1085 	return err ? ERR_PTR(err) : btf;
1086 }
1087 
1088 struct btf *btf__parse_raw(const char *path)
1089 {
1090 	return libbpf_ptr(btf_parse_raw(path, NULL));
1091 }
1092 
1093 struct btf *btf__parse_raw_split(const char *path, struct btf *base_btf)
1094 {
1095 	return libbpf_ptr(btf_parse_raw(path, base_btf));
1096 }
1097 
1098 static struct btf *btf_parse(const char *path, struct btf *base_btf, struct btf_ext **btf_ext)
1099 {
1100 	struct btf *btf;
1101 	int err;
1102 
1103 	if (btf_ext)
1104 		*btf_ext = NULL;
1105 
1106 	btf = btf_parse_raw(path, base_btf);
1107 	err = libbpf_get_error(btf);
1108 	if (!err)
1109 		return btf;
1110 	if (err != -EPROTO)
1111 		return ERR_PTR(err);
1112 	return btf_parse_elf(path, base_btf, btf_ext);
1113 }
1114 
1115 struct btf *btf__parse(const char *path, struct btf_ext **btf_ext)
1116 {
1117 	return libbpf_ptr(btf_parse(path, NULL, btf_ext));
1118 }
1119 
1120 struct btf *btf__parse_split(const char *path, struct btf *base_btf)
1121 {
1122 	return libbpf_ptr(btf_parse(path, base_btf, NULL));
1123 }
1124 
1125 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian);
1126 
1127 int btf_load_into_kernel(struct btf *btf, char *log_buf, size_t log_sz, __u32 log_level)
1128 {
1129 	LIBBPF_OPTS(bpf_btf_load_opts, opts);
1130 	__u32 buf_sz = 0, raw_size;
1131 	char *buf = NULL, *tmp;
1132 	void *raw_data;
1133 	int err = 0;
1134 
1135 	if (btf->fd >= 0)
1136 		return libbpf_err(-EEXIST);
1137 	if (log_sz && !log_buf)
1138 		return libbpf_err(-EINVAL);
1139 
1140 	/* cache native raw data representation */
1141 	raw_data = btf_get_raw_data(btf, &raw_size, false);
1142 	if (!raw_data) {
1143 		err = -ENOMEM;
1144 		goto done;
1145 	}
1146 	btf->raw_size = raw_size;
1147 	btf->raw_data = raw_data;
1148 
1149 retry_load:
1150 	/* if log_level is 0, we won't provide log_buf/log_size to the kernel,
1151 	 * initially. Only if BTF loading fails, we bump log_level to 1 and
1152 	 * retry, using either auto-allocated or custom log_buf. This way
1153 	 * non-NULL custom log_buf provides a buffer just in case, but hopes
1154 	 * for successful load and no need for log_buf.
1155 	 */
1156 	if (log_level) {
1157 		/* if caller didn't provide custom log_buf, we'll keep
1158 		 * allocating our own progressively bigger buffers for BTF
1159 		 * verification log
1160 		 */
1161 		if (!log_buf) {
1162 			buf_sz = max((__u32)BPF_LOG_BUF_SIZE, buf_sz * 2);
1163 			tmp = realloc(buf, buf_sz);
1164 			if (!tmp) {
1165 				err = -ENOMEM;
1166 				goto done;
1167 			}
1168 			buf = tmp;
1169 			buf[0] = '\0';
1170 		}
1171 
1172 		opts.log_buf = log_buf ? log_buf : buf;
1173 		opts.log_size = log_buf ? log_sz : buf_sz;
1174 		opts.log_level = log_level;
1175 	}
1176 
1177 	btf->fd = bpf_btf_load(raw_data, raw_size, &opts);
1178 	if (btf->fd < 0) {
1179 		/* time to turn on verbose mode and try again */
1180 		if (log_level == 0) {
1181 			log_level = 1;
1182 			goto retry_load;
1183 		}
1184 		/* only retry if caller didn't provide custom log_buf, but
1185 		 * make sure we can never overflow buf_sz
1186 		 */
1187 		if (!log_buf && errno == ENOSPC && buf_sz <= UINT_MAX / 2)
1188 			goto retry_load;
1189 
1190 		err = -errno;
1191 		pr_warn("BTF loading error: %d\n", err);
1192 		/* don't print out contents of custom log_buf */
1193 		if (!log_buf && buf[0])
1194 			pr_warn("-- BEGIN BTF LOAD LOG ---\n%s\n-- END BTF LOAD LOG --\n", buf);
1195 	}
1196 
1197 done:
1198 	free(buf);
1199 	return libbpf_err(err);
1200 }
1201 
1202 int btf__load_into_kernel(struct btf *btf)
1203 {
1204 	return btf_load_into_kernel(btf, NULL, 0, 0);
1205 }
1206 
1207 int btf__load(struct btf *) __attribute__((alias("btf__load_into_kernel")));
1208 
1209 int btf__fd(const struct btf *btf)
1210 {
1211 	return btf->fd;
1212 }
1213 
1214 void btf__set_fd(struct btf *btf, int fd)
1215 {
1216 	btf->fd = fd;
1217 }
1218 
1219 static const void *btf_strs_data(const struct btf *btf)
1220 {
1221 	return btf->strs_data ? btf->strs_data : strset__data(btf->strs_set);
1222 }
1223 
1224 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian)
1225 {
1226 	struct btf_header *hdr = btf->hdr;
1227 	struct btf_type *t;
1228 	void *data, *p;
1229 	__u32 data_sz;
1230 	int i;
1231 
1232 	data = swap_endian ? btf->raw_data_swapped : btf->raw_data;
1233 	if (data) {
1234 		*size = btf->raw_size;
1235 		return data;
1236 	}
1237 
1238 	data_sz = hdr->hdr_len + hdr->type_len + hdr->str_len;
1239 	data = calloc(1, data_sz);
1240 	if (!data)
1241 		return NULL;
1242 	p = data;
1243 
1244 	memcpy(p, hdr, hdr->hdr_len);
1245 	if (swap_endian)
1246 		btf_bswap_hdr(p);
1247 	p += hdr->hdr_len;
1248 
1249 	memcpy(p, btf->types_data, hdr->type_len);
1250 	if (swap_endian) {
1251 		for (i = 0; i < btf->nr_types; i++) {
1252 			t = p + btf->type_offs[i];
1253 			/* btf_bswap_type_rest() relies on native t->info, so
1254 			 * we swap base type info after we swapped all the
1255 			 * additional information
1256 			 */
1257 			if (btf_bswap_type_rest(t))
1258 				goto err_out;
1259 			btf_bswap_type_base(t);
1260 		}
1261 	}
1262 	p += hdr->type_len;
1263 
1264 	memcpy(p, btf_strs_data(btf), hdr->str_len);
1265 	p += hdr->str_len;
1266 
1267 	*size = data_sz;
1268 	return data;
1269 err_out:
1270 	free(data);
1271 	return NULL;
1272 }
1273 
1274 const void *btf__raw_data(const struct btf *btf_ro, __u32 *size)
1275 {
1276 	struct btf *btf = (struct btf *)btf_ro;
1277 	__u32 data_sz;
1278 	void *data;
1279 
1280 	data = btf_get_raw_data(btf, &data_sz, btf->swapped_endian);
1281 	if (!data)
1282 		return errno = ENOMEM, NULL;
1283 
1284 	btf->raw_size = data_sz;
1285 	if (btf->swapped_endian)
1286 		btf->raw_data_swapped = data;
1287 	else
1288 		btf->raw_data = data;
1289 	*size = data_sz;
1290 	return data;
1291 }
1292 
1293 __attribute__((alias("btf__raw_data")))
1294 const void *btf__get_raw_data(const struct btf *btf, __u32 *size);
1295 
1296 const char *btf__str_by_offset(const struct btf *btf, __u32 offset)
1297 {
1298 	if (offset < btf->start_str_off)
1299 		return btf__str_by_offset(btf->base_btf, offset);
1300 	else if (offset - btf->start_str_off < btf->hdr->str_len)
1301 		return btf_strs_data(btf) + (offset - btf->start_str_off);
1302 	else
1303 		return errno = EINVAL, NULL;
1304 }
1305 
1306 const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
1307 {
1308 	return btf__str_by_offset(btf, offset);
1309 }
1310 
1311 struct btf *btf_get_from_fd(int btf_fd, struct btf *base_btf)
1312 {
1313 	struct bpf_btf_info btf_info;
1314 	__u32 len = sizeof(btf_info);
1315 	__u32 last_size;
1316 	struct btf *btf;
1317 	void *ptr;
1318 	int err;
1319 
1320 	/* we won't know btf_size until we call bpf_obj_get_info_by_fd(). so
1321 	 * let's start with a sane default - 4KiB here - and resize it only if
1322 	 * bpf_obj_get_info_by_fd() needs a bigger buffer.
1323 	 */
1324 	last_size = 4096;
1325 	ptr = malloc(last_size);
1326 	if (!ptr)
1327 		return ERR_PTR(-ENOMEM);
1328 
1329 	memset(&btf_info, 0, sizeof(btf_info));
1330 	btf_info.btf = ptr_to_u64(ptr);
1331 	btf_info.btf_size = last_size;
1332 	err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1333 
1334 	if (!err && btf_info.btf_size > last_size) {
1335 		void *temp_ptr;
1336 
1337 		last_size = btf_info.btf_size;
1338 		temp_ptr = realloc(ptr, last_size);
1339 		if (!temp_ptr) {
1340 			btf = ERR_PTR(-ENOMEM);
1341 			goto exit_free;
1342 		}
1343 		ptr = temp_ptr;
1344 
1345 		len = sizeof(btf_info);
1346 		memset(&btf_info, 0, sizeof(btf_info));
1347 		btf_info.btf = ptr_to_u64(ptr);
1348 		btf_info.btf_size = last_size;
1349 
1350 		err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1351 	}
1352 
1353 	if (err || btf_info.btf_size > last_size) {
1354 		btf = err ? ERR_PTR(-errno) : ERR_PTR(-E2BIG);
1355 		goto exit_free;
1356 	}
1357 
1358 	btf = btf_new(ptr, btf_info.btf_size, base_btf);
1359 
1360 exit_free:
1361 	free(ptr);
1362 	return btf;
1363 }
1364 
1365 struct btf *btf__load_from_kernel_by_id_split(__u32 id, struct btf *base_btf)
1366 {
1367 	struct btf *btf;
1368 	int btf_fd;
1369 
1370 	btf_fd = bpf_btf_get_fd_by_id(id);
1371 	if (btf_fd < 0)
1372 		return libbpf_err_ptr(-errno);
1373 
1374 	btf = btf_get_from_fd(btf_fd, base_btf);
1375 	close(btf_fd);
1376 
1377 	return libbpf_ptr(btf);
1378 }
1379 
1380 struct btf *btf__load_from_kernel_by_id(__u32 id)
1381 {
1382 	return btf__load_from_kernel_by_id_split(id, NULL);
1383 }
1384 
1385 int btf__get_from_id(__u32 id, struct btf **btf)
1386 {
1387 	struct btf *res;
1388 	int err;
1389 
1390 	*btf = NULL;
1391 	res = btf__load_from_kernel_by_id(id);
1392 	err = libbpf_get_error(res);
1393 
1394 	if (err)
1395 		return libbpf_err(err);
1396 
1397 	*btf = res;
1398 	return 0;
1399 }
1400 
1401 int btf__get_map_kv_tids(const struct btf *btf, const char *map_name,
1402 			 __u32 expected_key_size, __u32 expected_value_size,
1403 			 __u32 *key_type_id, __u32 *value_type_id)
1404 {
1405 	const struct btf_type *container_type;
1406 	const struct btf_member *key, *value;
1407 	const size_t max_name = 256;
1408 	char container_name[max_name];
1409 	__s64 key_size, value_size;
1410 	__s32 container_id;
1411 
1412 	if (snprintf(container_name, max_name, "____btf_map_%s", map_name) == max_name) {
1413 		pr_warn("map:%s length of '____btf_map_%s' is too long\n",
1414 			map_name, map_name);
1415 		return libbpf_err(-EINVAL);
1416 	}
1417 
1418 	container_id = btf__find_by_name(btf, container_name);
1419 	if (container_id < 0) {
1420 		pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
1421 			 map_name, container_name);
1422 		return libbpf_err(container_id);
1423 	}
1424 
1425 	container_type = btf__type_by_id(btf, container_id);
1426 	if (!container_type) {
1427 		pr_warn("map:%s cannot find BTF type for container_id:%u\n",
1428 			map_name, container_id);
1429 		return libbpf_err(-EINVAL);
1430 	}
1431 
1432 	if (!btf_is_struct(container_type) || btf_vlen(container_type) < 2) {
1433 		pr_warn("map:%s container_name:%s is an invalid container struct\n",
1434 			map_name, container_name);
1435 		return libbpf_err(-EINVAL);
1436 	}
1437 
1438 	key = btf_members(container_type);
1439 	value = key + 1;
1440 
1441 	key_size = btf__resolve_size(btf, key->type);
1442 	if (key_size < 0) {
1443 		pr_warn("map:%s invalid BTF key_type_size\n", map_name);
1444 		return libbpf_err(key_size);
1445 	}
1446 
1447 	if (expected_key_size != key_size) {
1448 		pr_warn("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
1449 			map_name, (__u32)key_size, expected_key_size);
1450 		return libbpf_err(-EINVAL);
1451 	}
1452 
1453 	value_size = btf__resolve_size(btf, value->type);
1454 	if (value_size < 0) {
1455 		pr_warn("map:%s invalid BTF value_type_size\n", map_name);
1456 		return libbpf_err(value_size);
1457 	}
1458 
1459 	if (expected_value_size != value_size) {
1460 		pr_warn("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
1461 			map_name, (__u32)value_size, expected_value_size);
1462 		return libbpf_err(-EINVAL);
1463 	}
1464 
1465 	*key_type_id = key->type;
1466 	*value_type_id = value->type;
1467 
1468 	return 0;
1469 }
1470 
1471 static void btf_invalidate_raw_data(struct btf *btf)
1472 {
1473 	if (btf->raw_data) {
1474 		free(btf->raw_data);
1475 		btf->raw_data = NULL;
1476 	}
1477 	if (btf->raw_data_swapped) {
1478 		free(btf->raw_data_swapped);
1479 		btf->raw_data_swapped = NULL;
1480 	}
1481 }
1482 
1483 /* Ensure BTF is ready to be modified (by splitting into a three memory
1484  * regions for header, types, and strings). Also invalidate cached
1485  * raw_data, if any.
1486  */
1487 static int btf_ensure_modifiable(struct btf *btf)
1488 {
1489 	void *hdr, *types;
1490 	struct strset *set = NULL;
1491 	int err = -ENOMEM;
1492 
1493 	if (btf_is_modifiable(btf)) {
1494 		/* any BTF modification invalidates raw_data */
1495 		btf_invalidate_raw_data(btf);
1496 		return 0;
1497 	}
1498 
1499 	/* split raw data into three memory regions */
1500 	hdr = malloc(btf->hdr->hdr_len);
1501 	types = malloc(btf->hdr->type_len);
1502 	if (!hdr || !types)
1503 		goto err_out;
1504 
1505 	memcpy(hdr, btf->hdr, btf->hdr->hdr_len);
1506 	memcpy(types, btf->types_data, btf->hdr->type_len);
1507 
1508 	/* build lookup index for all strings */
1509 	set = strset__new(BTF_MAX_STR_OFFSET, btf->strs_data, btf->hdr->str_len);
1510 	if (IS_ERR(set)) {
1511 		err = PTR_ERR(set);
1512 		goto err_out;
1513 	}
1514 
1515 	/* only when everything was successful, update internal state */
1516 	btf->hdr = hdr;
1517 	btf->types_data = types;
1518 	btf->types_data_cap = btf->hdr->type_len;
1519 	btf->strs_data = NULL;
1520 	btf->strs_set = set;
1521 	/* if BTF was created from scratch, all strings are guaranteed to be
1522 	 * unique and deduplicated
1523 	 */
1524 	if (btf->hdr->str_len == 0)
1525 		btf->strs_deduped = true;
1526 	if (!btf->base_btf && btf->hdr->str_len == 1)
1527 		btf->strs_deduped = true;
1528 
1529 	/* invalidate raw_data representation */
1530 	btf_invalidate_raw_data(btf);
1531 
1532 	return 0;
1533 
1534 err_out:
1535 	strset__free(set);
1536 	free(hdr);
1537 	free(types);
1538 	return err;
1539 }
1540 
1541 /* Find an offset in BTF string section that corresponds to a given string *s*.
1542  * Returns:
1543  *   - >0 offset into string section, if string is found;
1544  *   - -ENOENT, if string is not in the string section;
1545  *   - <0, on any other error.
1546  */
1547 int btf__find_str(struct btf *btf, const char *s)
1548 {
1549 	int off;
1550 
1551 	if (btf->base_btf) {
1552 		off = btf__find_str(btf->base_btf, s);
1553 		if (off != -ENOENT)
1554 			return off;
1555 	}
1556 
1557 	/* BTF needs to be in a modifiable state to build string lookup index */
1558 	if (btf_ensure_modifiable(btf))
1559 		return libbpf_err(-ENOMEM);
1560 
1561 	off = strset__find_str(btf->strs_set, s);
1562 	if (off < 0)
1563 		return libbpf_err(off);
1564 
1565 	return btf->start_str_off + off;
1566 }
1567 
1568 /* Add a string s to the BTF string section.
1569  * Returns:
1570  *   - > 0 offset into string section, on success;
1571  *   - < 0, on error.
1572  */
1573 int btf__add_str(struct btf *btf, const char *s)
1574 {
1575 	int off;
1576 
1577 	if (btf->base_btf) {
1578 		off = btf__find_str(btf->base_btf, s);
1579 		if (off != -ENOENT)
1580 			return off;
1581 	}
1582 
1583 	if (btf_ensure_modifiable(btf))
1584 		return libbpf_err(-ENOMEM);
1585 
1586 	off = strset__add_str(btf->strs_set, s);
1587 	if (off < 0)
1588 		return libbpf_err(off);
1589 
1590 	btf->hdr->str_len = strset__data_size(btf->strs_set);
1591 
1592 	return btf->start_str_off + off;
1593 }
1594 
1595 static void *btf_add_type_mem(struct btf *btf, size_t add_sz)
1596 {
1597 	return libbpf_add_mem(&btf->types_data, &btf->types_data_cap, 1,
1598 			      btf->hdr->type_len, UINT_MAX, add_sz);
1599 }
1600 
1601 static void btf_type_inc_vlen(struct btf_type *t)
1602 {
1603 	t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, btf_kflag(t));
1604 }
1605 
1606 static int btf_commit_type(struct btf *btf, int data_sz)
1607 {
1608 	int err;
1609 
1610 	err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1611 	if (err)
1612 		return libbpf_err(err);
1613 
1614 	btf->hdr->type_len += data_sz;
1615 	btf->hdr->str_off += data_sz;
1616 	btf->nr_types++;
1617 	return btf->start_id + btf->nr_types - 1;
1618 }
1619 
1620 struct btf_pipe {
1621 	const struct btf *src;
1622 	struct btf *dst;
1623 };
1624 
1625 static int btf_rewrite_str(__u32 *str_off, void *ctx)
1626 {
1627 	struct btf_pipe *p = ctx;
1628 	int off;
1629 
1630 	if (!*str_off) /* nothing to do for empty strings */
1631 		return 0;
1632 
1633 	off = btf__add_str(p->dst, btf__str_by_offset(p->src, *str_off));
1634 	if (off < 0)
1635 		return off;
1636 
1637 	*str_off = off;
1638 	return 0;
1639 }
1640 
1641 int btf__add_type(struct btf *btf, const struct btf *src_btf, const struct btf_type *src_type)
1642 {
1643 	struct btf_pipe p = { .src = src_btf, .dst = btf };
1644 	struct btf_type *t;
1645 	int sz, err;
1646 
1647 	sz = btf_type_size(src_type);
1648 	if (sz < 0)
1649 		return libbpf_err(sz);
1650 
1651 	/* deconstruct BTF, if necessary, and invalidate raw_data */
1652 	if (btf_ensure_modifiable(btf))
1653 		return libbpf_err(-ENOMEM);
1654 
1655 	t = btf_add_type_mem(btf, sz);
1656 	if (!t)
1657 		return libbpf_err(-ENOMEM);
1658 
1659 	memcpy(t, src_type, sz);
1660 
1661 	err = btf_type_visit_str_offs(t, btf_rewrite_str, &p);
1662 	if (err)
1663 		return libbpf_err(err);
1664 
1665 	return btf_commit_type(btf, sz);
1666 }
1667 
1668 static int btf_rewrite_type_ids(__u32 *type_id, void *ctx)
1669 {
1670 	struct btf *btf = ctx;
1671 
1672 	if (!*type_id) /* nothing to do for VOID references */
1673 		return 0;
1674 
1675 	/* we haven't updated btf's type count yet, so
1676 	 * btf->start_id + btf->nr_types - 1 is the type ID offset we should
1677 	 * add to all newly added BTF types
1678 	 */
1679 	*type_id += btf->start_id + btf->nr_types - 1;
1680 	return 0;
1681 }
1682 
1683 int btf__add_btf(struct btf *btf, const struct btf *src_btf)
1684 {
1685 	struct btf_pipe p = { .src = src_btf, .dst = btf };
1686 	int data_sz, sz, cnt, i, err, old_strs_len;
1687 	__u32 *off;
1688 	void *t;
1689 
1690 	/* appending split BTF isn't supported yet */
1691 	if (src_btf->base_btf)
1692 		return libbpf_err(-ENOTSUP);
1693 
1694 	/* deconstruct BTF, if necessary, and invalidate raw_data */
1695 	if (btf_ensure_modifiable(btf))
1696 		return libbpf_err(-ENOMEM);
1697 
1698 	/* remember original strings section size if we have to roll back
1699 	 * partial strings section changes
1700 	 */
1701 	old_strs_len = btf->hdr->str_len;
1702 
1703 	data_sz = src_btf->hdr->type_len;
1704 	cnt = btf__type_cnt(src_btf) - 1;
1705 
1706 	/* pre-allocate enough memory for new types */
1707 	t = btf_add_type_mem(btf, data_sz);
1708 	if (!t)
1709 		return libbpf_err(-ENOMEM);
1710 
1711 	/* pre-allocate enough memory for type offset index for new types */
1712 	off = btf_add_type_offs_mem(btf, cnt);
1713 	if (!off)
1714 		return libbpf_err(-ENOMEM);
1715 
1716 	/* bulk copy types data for all types from src_btf */
1717 	memcpy(t, src_btf->types_data, data_sz);
1718 
1719 	for (i = 0; i < cnt; i++) {
1720 		sz = btf_type_size(t);
1721 		if (sz < 0) {
1722 			/* unlikely, has to be corrupted src_btf */
1723 			err = sz;
1724 			goto err_out;
1725 		}
1726 
1727 		/* fill out type ID to type offset mapping for lookups by type ID */
1728 		*off = t - btf->types_data;
1729 
1730 		/* add, dedup, and remap strings referenced by this BTF type */
1731 		err = btf_type_visit_str_offs(t, btf_rewrite_str, &p);
1732 		if (err)
1733 			goto err_out;
1734 
1735 		/* remap all type IDs referenced from this BTF type */
1736 		err = btf_type_visit_type_ids(t, btf_rewrite_type_ids, btf);
1737 		if (err)
1738 			goto err_out;
1739 
1740 		/* go to next type data and type offset index entry */
1741 		t += sz;
1742 		off++;
1743 	}
1744 
1745 	/* Up until now any of the copied type data was effectively invisible,
1746 	 * so if we exited early before this point due to error, BTF would be
1747 	 * effectively unmodified. There would be extra internal memory
1748 	 * pre-allocated, but it would not be available for querying.  But now
1749 	 * that we've copied and rewritten all the data successfully, we can
1750 	 * update type count and various internal offsets and sizes to
1751 	 * "commit" the changes and made them visible to the outside world.
1752 	 */
1753 	btf->hdr->type_len += data_sz;
1754 	btf->hdr->str_off += data_sz;
1755 	btf->nr_types += cnt;
1756 
1757 	/* return type ID of the first added BTF type */
1758 	return btf->start_id + btf->nr_types - cnt;
1759 err_out:
1760 	/* zero out preallocated memory as if it was just allocated with
1761 	 * libbpf_add_mem()
1762 	 */
1763 	memset(btf->types_data + btf->hdr->type_len, 0, data_sz);
1764 	memset(btf->strs_data + old_strs_len, 0, btf->hdr->str_len - old_strs_len);
1765 
1766 	/* and now restore original strings section size; types data size
1767 	 * wasn't modified, so doesn't need restoring, see big comment above */
1768 	btf->hdr->str_len = old_strs_len;
1769 
1770 	return libbpf_err(err);
1771 }
1772 
1773 /*
1774  * Append new BTF_KIND_INT type with:
1775  *   - *name* - non-empty, non-NULL type name;
1776  *   - *sz* - power-of-2 (1, 2, 4, ..) size of the type, in bytes;
1777  *   - encoding is a combination of BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL.
1778  * Returns:
1779  *   - >0, type ID of newly added BTF type;
1780  *   - <0, on error.
1781  */
1782 int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding)
1783 {
1784 	struct btf_type *t;
1785 	int sz, name_off;
1786 
1787 	/* non-empty name */
1788 	if (!name || !name[0])
1789 		return libbpf_err(-EINVAL);
1790 	/* byte_sz must be power of 2 */
1791 	if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
1792 		return libbpf_err(-EINVAL);
1793 	if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
1794 		return libbpf_err(-EINVAL);
1795 
1796 	/* deconstruct BTF, if necessary, and invalidate raw_data */
1797 	if (btf_ensure_modifiable(btf))
1798 		return libbpf_err(-ENOMEM);
1799 
1800 	sz = sizeof(struct btf_type) + sizeof(int);
1801 	t = btf_add_type_mem(btf, sz);
1802 	if (!t)
1803 		return libbpf_err(-ENOMEM);
1804 
1805 	/* if something goes wrong later, we might end up with an extra string,
1806 	 * but that shouldn't be a problem, because BTF can't be constructed
1807 	 * completely anyway and will most probably be just discarded
1808 	 */
1809 	name_off = btf__add_str(btf, name);
1810 	if (name_off < 0)
1811 		return name_off;
1812 
1813 	t->name_off = name_off;
1814 	t->info = btf_type_info(BTF_KIND_INT, 0, 0);
1815 	t->size = byte_sz;
1816 	/* set INT info, we don't allow setting legacy bit offset/size */
1817 	*(__u32 *)(t + 1) = (encoding << 24) | (byte_sz * 8);
1818 
1819 	return btf_commit_type(btf, sz);
1820 }
1821 
1822 /*
1823  * Append new BTF_KIND_FLOAT type with:
1824  *   - *name* - non-empty, non-NULL type name;
1825  *   - *sz* - size of the type, in bytes;
1826  * Returns:
1827  *   - >0, type ID of newly added BTF type;
1828  *   - <0, on error.
1829  */
1830 int btf__add_float(struct btf *btf, const char *name, size_t byte_sz)
1831 {
1832 	struct btf_type *t;
1833 	int sz, name_off;
1834 
1835 	/* non-empty name */
1836 	if (!name || !name[0])
1837 		return libbpf_err(-EINVAL);
1838 
1839 	/* byte_sz must be one of the explicitly allowed values */
1840 	if (byte_sz != 2 && byte_sz != 4 && byte_sz != 8 && byte_sz != 12 &&
1841 	    byte_sz != 16)
1842 		return libbpf_err(-EINVAL);
1843 
1844 	if (btf_ensure_modifiable(btf))
1845 		return libbpf_err(-ENOMEM);
1846 
1847 	sz = sizeof(struct btf_type);
1848 	t = btf_add_type_mem(btf, sz);
1849 	if (!t)
1850 		return libbpf_err(-ENOMEM);
1851 
1852 	name_off = btf__add_str(btf, name);
1853 	if (name_off < 0)
1854 		return name_off;
1855 
1856 	t->name_off = name_off;
1857 	t->info = btf_type_info(BTF_KIND_FLOAT, 0, 0);
1858 	t->size = byte_sz;
1859 
1860 	return btf_commit_type(btf, sz);
1861 }
1862 
1863 /* it's completely legal to append BTF types with type IDs pointing forward to
1864  * types that haven't been appended yet, so we only make sure that id looks
1865  * sane, we can't guarantee that ID will always be valid
1866  */
1867 static int validate_type_id(int id)
1868 {
1869 	if (id < 0 || id > BTF_MAX_NR_TYPES)
1870 		return -EINVAL;
1871 	return 0;
1872 }
1873 
1874 /* generic append function for PTR, TYPEDEF, CONST/VOLATILE/RESTRICT */
1875 static int btf_add_ref_kind(struct btf *btf, int kind, const char *name, int ref_type_id)
1876 {
1877 	struct btf_type *t;
1878 	int sz, name_off = 0;
1879 
1880 	if (validate_type_id(ref_type_id))
1881 		return libbpf_err(-EINVAL);
1882 
1883 	if (btf_ensure_modifiable(btf))
1884 		return libbpf_err(-ENOMEM);
1885 
1886 	sz = sizeof(struct btf_type);
1887 	t = btf_add_type_mem(btf, sz);
1888 	if (!t)
1889 		return libbpf_err(-ENOMEM);
1890 
1891 	if (name && name[0]) {
1892 		name_off = btf__add_str(btf, name);
1893 		if (name_off < 0)
1894 			return name_off;
1895 	}
1896 
1897 	t->name_off = name_off;
1898 	t->info = btf_type_info(kind, 0, 0);
1899 	t->type = ref_type_id;
1900 
1901 	return btf_commit_type(btf, sz);
1902 }
1903 
1904 /*
1905  * Append new BTF_KIND_PTR type with:
1906  *   - *ref_type_id* - referenced type ID, it might not exist yet;
1907  * Returns:
1908  *   - >0, type ID of newly added BTF type;
1909  *   - <0, on error.
1910  */
1911 int btf__add_ptr(struct btf *btf, int ref_type_id)
1912 {
1913 	return btf_add_ref_kind(btf, BTF_KIND_PTR, NULL, ref_type_id);
1914 }
1915 
1916 /*
1917  * Append new BTF_KIND_ARRAY type with:
1918  *   - *index_type_id* - type ID of the type describing array index;
1919  *   - *elem_type_id* - type ID of the type describing array element;
1920  *   - *nr_elems* - the size of the array;
1921  * Returns:
1922  *   - >0, type ID of newly added BTF type;
1923  *   - <0, on error.
1924  */
1925 int btf__add_array(struct btf *btf, int index_type_id, int elem_type_id, __u32 nr_elems)
1926 {
1927 	struct btf_type *t;
1928 	struct btf_array *a;
1929 	int sz;
1930 
1931 	if (validate_type_id(index_type_id) || validate_type_id(elem_type_id))
1932 		return libbpf_err(-EINVAL);
1933 
1934 	if (btf_ensure_modifiable(btf))
1935 		return libbpf_err(-ENOMEM);
1936 
1937 	sz = sizeof(struct btf_type) + sizeof(struct btf_array);
1938 	t = btf_add_type_mem(btf, sz);
1939 	if (!t)
1940 		return libbpf_err(-ENOMEM);
1941 
1942 	t->name_off = 0;
1943 	t->info = btf_type_info(BTF_KIND_ARRAY, 0, 0);
1944 	t->size = 0;
1945 
1946 	a = btf_array(t);
1947 	a->type = elem_type_id;
1948 	a->index_type = index_type_id;
1949 	a->nelems = nr_elems;
1950 
1951 	return btf_commit_type(btf, sz);
1952 }
1953 
1954 /* generic STRUCT/UNION append function */
1955 static int btf_add_composite(struct btf *btf, int kind, const char *name, __u32 bytes_sz)
1956 {
1957 	struct btf_type *t;
1958 	int sz, name_off = 0;
1959 
1960 	if (btf_ensure_modifiable(btf))
1961 		return libbpf_err(-ENOMEM);
1962 
1963 	sz = sizeof(struct btf_type);
1964 	t = btf_add_type_mem(btf, sz);
1965 	if (!t)
1966 		return libbpf_err(-ENOMEM);
1967 
1968 	if (name && name[0]) {
1969 		name_off = btf__add_str(btf, name);
1970 		if (name_off < 0)
1971 			return name_off;
1972 	}
1973 
1974 	/* start out with vlen=0 and no kflag; this will be adjusted when
1975 	 * adding each member
1976 	 */
1977 	t->name_off = name_off;
1978 	t->info = btf_type_info(kind, 0, 0);
1979 	t->size = bytes_sz;
1980 
1981 	return btf_commit_type(btf, sz);
1982 }
1983 
1984 /*
1985  * Append new BTF_KIND_STRUCT type with:
1986  *   - *name* - name of the struct, can be NULL or empty for anonymous structs;
1987  *   - *byte_sz* - size of the struct, in bytes;
1988  *
1989  * Struct initially has no fields in it. Fields can be added by
1990  * btf__add_field() right after btf__add_struct() succeeds.
1991  *
1992  * Returns:
1993  *   - >0, type ID of newly added BTF type;
1994  *   - <0, on error.
1995  */
1996 int btf__add_struct(struct btf *btf, const char *name, __u32 byte_sz)
1997 {
1998 	return btf_add_composite(btf, BTF_KIND_STRUCT, name, byte_sz);
1999 }
2000 
2001 /*
2002  * Append new BTF_KIND_UNION type with:
2003  *   - *name* - name of the union, can be NULL or empty for anonymous union;
2004  *   - *byte_sz* - size of the union, in bytes;
2005  *
2006  * Union initially has no fields in it. Fields can be added by
2007  * btf__add_field() right after btf__add_union() succeeds. All fields
2008  * should have *bit_offset* of 0.
2009  *
2010  * Returns:
2011  *   - >0, type ID of newly added BTF type;
2012  *   - <0, on error.
2013  */
2014 int btf__add_union(struct btf *btf, const char *name, __u32 byte_sz)
2015 {
2016 	return btf_add_composite(btf, BTF_KIND_UNION, name, byte_sz);
2017 }
2018 
2019 static struct btf_type *btf_last_type(struct btf *btf)
2020 {
2021 	return btf_type_by_id(btf, btf__type_cnt(btf) - 1);
2022 }
2023 
2024 /*
2025  * Append new field for the current STRUCT/UNION type with:
2026  *   - *name* - name of the field, can be NULL or empty for anonymous field;
2027  *   - *type_id* - type ID for the type describing field type;
2028  *   - *bit_offset* - bit offset of the start of the field within struct/union;
2029  *   - *bit_size* - bit size of a bitfield, 0 for non-bitfield fields;
2030  * Returns:
2031  *   -  0, on success;
2032  *   - <0, on error.
2033  */
2034 int btf__add_field(struct btf *btf, const char *name, int type_id,
2035 		   __u32 bit_offset, __u32 bit_size)
2036 {
2037 	struct btf_type *t;
2038 	struct btf_member *m;
2039 	bool is_bitfield;
2040 	int sz, name_off = 0;
2041 
2042 	/* last type should be union/struct */
2043 	if (btf->nr_types == 0)
2044 		return libbpf_err(-EINVAL);
2045 	t = btf_last_type(btf);
2046 	if (!btf_is_composite(t))
2047 		return libbpf_err(-EINVAL);
2048 
2049 	if (validate_type_id(type_id))
2050 		return libbpf_err(-EINVAL);
2051 	/* best-effort bit field offset/size enforcement */
2052 	is_bitfield = bit_size || (bit_offset % 8 != 0);
2053 	if (is_bitfield && (bit_size == 0 || bit_size > 255 || bit_offset > 0xffffff))
2054 		return libbpf_err(-EINVAL);
2055 
2056 	/* only offset 0 is allowed for unions */
2057 	if (btf_is_union(t) && bit_offset)
2058 		return libbpf_err(-EINVAL);
2059 
2060 	/* decompose and invalidate raw data */
2061 	if (btf_ensure_modifiable(btf))
2062 		return libbpf_err(-ENOMEM);
2063 
2064 	sz = sizeof(struct btf_member);
2065 	m = btf_add_type_mem(btf, sz);
2066 	if (!m)
2067 		return libbpf_err(-ENOMEM);
2068 
2069 	if (name && name[0]) {
2070 		name_off = btf__add_str(btf, name);
2071 		if (name_off < 0)
2072 			return name_off;
2073 	}
2074 
2075 	m->name_off = name_off;
2076 	m->type = type_id;
2077 	m->offset = bit_offset | (bit_size << 24);
2078 
2079 	/* btf_add_type_mem can invalidate t pointer */
2080 	t = btf_last_type(btf);
2081 	/* update parent type's vlen and kflag */
2082 	t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, is_bitfield || btf_kflag(t));
2083 
2084 	btf->hdr->type_len += sz;
2085 	btf->hdr->str_off += sz;
2086 	return 0;
2087 }
2088 
2089 /*
2090  * Append new BTF_KIND_ENUM type with:
2091  *   - *name* - name of the enum, can be NULL or empty for anonymous enums;
2092  *   - *byte_sz* - size of the enum, in bytes.
2093  *
2094  * Enum initially has no enum values in it (and corresponds to enum forward
2095  * declaration). Enumerator values can be added by btf__add_enum_value()
2096  * immediately after btf__add_enum() succeeds.
2097  *
2098  * Returns:
2099  *   - >0, type ID of newly added BTF type;
2100  *   - <0, on error.
2101  */
2102 int btf__add_enum(struct btf *btf, const char *name, __u32 byte_sz)
2103 {
2104 	struct btf_type *t;
2105 	int sz, name_off = 0;
2106 
2107 	/* byte_sz must be power of 2 */
2108 	if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 8)
2109 		return libbpf_err(-EINVAL);
2110 
2111 	if (btf_ensure_modifiable(btf))
2112 		return libbpf_err(-ENOMEM);
2113 
2114 	sz = sizeof(struct btf_type);
2115 	t = btf_add_type_mem(btf, sz);
2116 	if (!t)
2117 		return libbpf_err(-ENOMEM);
2118 
2119 	if (name && name[0]) {
2120 		name_off = btf__add_str(btf, name);
2121 		if (name_off < 0)
2122 			return name_off;
2123 	}
2124 
2125 	/* start out with vlen=0; it will be adjusted when adding enum values */
2126 	t->name_off = name_off;
2127 	t->info = btf_type_info(BTF_KIND_ENUM, 0, 0);
2128 	t->size = byte_sz;
2129 
2130 	return btf_commit_type(btf, sz);
2131 }
2132 
2133 /*
2134  * Append new enum value for the current ENUM type with:
2135  *   - *name* - name of the enumerator value, can't be NULL or empty;
2136  *   - *value* - integer value corresponding to enum value *name*;
2137  * Returns:
2138  *   -  0, on success;
2139  *   - <0, on error.
2140  */
2141 int btf__add_enum_value(struct btf *btf, const char *name, __s64 value)
2142 {
2143 	struct btf_type *t;
2144 	struct btf_enum *v;
2145 	int sz, name_off;
2146 
2147 	/* last type should be BTF_KIND_ENUM */
2148 	if (btf->nr_types == 0)
2149 		return libbpf_err(-EINVAL);
2150 	t = btf_last_type(btf);
2151 	if (!btf_is_enum(t))
2152 		return libbpf_err(-EINVAL);
2153 
2154 	/* non-empty name */
2155 	if (!name || !name[0])
2156 		return libbpf_err(-EINVAL);
2157 	if (value < INT_MIN || value > UINT_MAX)
2158 		return libbpf_err(-E2BIG);
2159 
2160 	/* decompose and invalidate raw data */
2161 	if (btf_ensure_modifiable(btf))
2162 		return libbpf_err(-ENOMEM);
2163 
2164 	sz = sizeof(struct btf_enum);
2165 	v = btf_add_type_mem(btf, sz);
2166 	if (!v)
2167 		return libbpf_err(-ENOMEM);
2168 
2169 	name_off = btf__add_str(btf, name);
2170 	if (name_off < 0)
2171 		return name_off;
2172 
2173 	v->name_off = name_off;
2174 	v->val = value;
2175 
2176 	/* update parent type's vlen */
2177 	t = btf_last_type(btf);
2178 	btf_type_inc_vlen(t);
2179 
2180 	btf->hdr->type_len += sz;
2181 	btf->hdr->str_off += sz;
2182 	return 0;
2183 }
2184 
2185 /*
2186  * Append new BTF_KIND_FWD type with:
2187  *   - *name*, non-empty/non-NULL name;
2188  *   - *fwd_kind*, kind of forward declaration, one of BTF_FWD_STRUCT,
2189  *     BTF_FWD_UNION, or BTF_FWD_ENUM;
2190  * Returns:
2191  *   - >0, type ID of newly added BTF type;
2192  *   - <0, on error.
2193  */
2194 int btf__add_fwd(struct btf *btf, const char *name, enum btf_fwd_kind fwd_kind)
2195 {
2196 	if (!name || !name[0])
2197 		return libbpf_err(-EINVAL);
2198 
2199 	switch (fwd_kind) {
2200 	case BTF_FWD_STRUCT:
2201 	case BTF_FWD_UNION: {
2202 		struct btf_type *t;
2203 		int id;
2204 
2205 		id = btf_add_ref_kind(btf, BTF_KIND_FWD, name, 0);
2206 		if (id <= 0)
2207 			return id;
2208 		t = btf_type_by_id(btf, id);
2209 		t->info = btf_type_info(BTF_KIND_FWD, 0, fwd_kind == BTF_FWD_UNION);
2210 		return id;
2211 	}
2212 	case BTF_FWD_ENUM:
2213 		/* enum forward in BTF currently is just an enum with no enum
2214 		 * values; we also assume a standard 4-byte size for it
2215 		 */
2216 		return btf__add_enum(btf, name, sizeof(int));
2217 	default:
2218 		return libbpf_err(-EINVAL);
2219 	}
2220 }
2221 
2222 /*
2223  * Append new BTF_KING_TYPEDEF type with:
2224  *   - *name*, non-empty/non-NULL name;
2225  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2226  * Returns:
2227  *   - >0, type ID of newly added BTF type;
2228  *   - <0, on error.
2229  */
2230 int btf__add_typedef(struct btf *btf, const char *name, int ref_type_id)
2231 {
2232 	if (!name || !name[0])
2233 		return libbpf_err(-EINVAL);
2234 
2235 	return btf_add_ref_kind(btf, BTF_KIND_TYPEDEF, name, ref_type_id);
2236 }
2237 
2238 /*
2239  * Append new BTF_KIND_VOLATILE type with:
2240  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2241  * Returns:
2242  *   - >0, type ID of newly added BTF type;
2243  *   - <0, on error.
2244  */
2245 int btf__add_volatile(struct btf *btf, int ref_type_id)
2246 {
2247 	return btf_add_ref_kind(btf, BTF_KIND_VOLATILE, NULL, ref_type_id);
2248 }
2249 
2250 /*
2251  * Append new BTF_KIND_CONST type with:
2252  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2253  * Returns:
2254  *   - >0, type ID of newly added BTF type;
2255  *   - <0, on error.
2256  */
2257 int btf__add_const(struct btf *btf, int ref_type_id)
2258 {
2259 	return btf_add_ref_kind(btf, BTF_KIND_CONST, NULL, ref_type_id);
2260 }
2261 
2262 /*
2263  * Append new BTF_KIND_RESTRICT type with:
2264  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2265  * Returns:
2266  *   - >0, type ID of newly added BTF type;
2267  *   - <0, on error.
2268  */
2269 int btf__add_restrict(struct btf *btf, int ref_type_id)
2270 {
2271 	return btf_add_ref_kind(btf, BTF_KIND_RESTRICT, NULL, ref_type_id);
2272 }
2273 
2274 /*
2275  * Append new BTF_KIND_TYPE_TAG type with:
2276  *   - *value*, non-empty/non-NULL tag value;
2277  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2278  * Returns:
2279  *   - >0, type ID of newly added BTF type;
2280  *   - <0, on error.
2281  */
2282 int btf__add_type_tag(struct btf *btf, const char *value, int ref_type_id)
2283 {
2284 	if (!value|| !value[0])
2285 		return libbpf_err(-EINVAL);
2286 
2287 	return btf_add_ref_kind(btf, BTF_KIND_TYPE_TAG, value, ref_type_id);
2288 }
2289 
2290 /*
2291  * Append new BTF_KIND_FUNC type with:
2292  *   - *name*, non-empty/non-NULL name;
2293  *   - *proto_type_id* - FUNC_PROTO's type ID, it might not exist yet;
2294  * Returns:
2295  *   - >0, type ID of newly added BTF type;
2296  *   - <0, on error.
2297  */
2298 int btf__add_func(struct btf *btf, const char *name,
2299 		  enum btf_func_linkage linkage, int proto_type_id)
2300 {
2301 	int id;
2302 
2303 	if (!name || !name[0])
2304 		return libbpf_err(-EINVAL);
2305 	if (linkage != BTF_FUNC_STATIC && linkage != BTF_FUNC_GLOBAL &&
2306 	    linkage != BTF_FUNC_EXTERN)
2307 		return libbpf_err(-EINVAL);
2308 
2309 	id = btf_add_ref_kind(btf, BTF_KIND_FUNC, name, proto_type_id);
2310 	if (id > 0) {
2311 		struct btf_type *t = btf_type_by_id(btf, id);
2312 
2313 		t->info = btf_type_info(BTF_KIND_FUNC, linkage, 0);
2314 	}
2315 	return libbpf_err(id);
2316 }
2317 
2318 /*
2319  * Append new BTF_KIND_FUNC_PROTO with:
2320  *   - *ret_type_id* - type ID for return result of a function.
2321  *
2322  * Function prototype initially has no arguments, but they can be added by
2323  * btf__add_func_param() one by one, immediately after
2324  * btf__add_func_proto() succeeded.
2325  *
2326  * Returns:
2327  *   - >0, type ID of newly added BTF type;
2328  *   - <0, on error.
2329  */
2330 int btf__add_func_proto(struct btf *btf, int ret_type_id)
2331 {
2332 	struct btf_type *t;
2333 	int sz;
2334 
2335 	if (validate_type_id(ret_type_id))
2336 		return libbpf_err(-EINVAL);
2337 
2338 	if (btf_ensure_modifiable(btf))
2339 		return libbpf_err(-ENOMEM);
2340 
2341 	sz = sizeof(struct btf_type);
2342 	t = btf_add_type_mem(btf, sz);
2343 	if (!t)
2344 		return libbpf_err(-ENOMEM);
2345 
2346 	/* start out with vlen=0; this will be adjusted when adding enum
2347 	 * values, if necessary
2348 	 */
2349 	t->name_off = 0;
2350 	t->info = btf_type_info(BTF_KIND_FUNC_PROTO, 0, 0);
2351 	t->type = ret_type_id;
2352 
2353 	return btf_commit_type(btf, sz);
2354 }
2355 
2356 /*
2357  * Append new function parameter for current FUNC_PROTO type with:
2358  *   - *name* - parameter name, can be NULL or empty;
2359  *   - *type_id* - type ID describing the type of the parameter.
2360  * Returns:
2361  *   -  0, on success;
2362  *   - <0, on error.
2363  */
2364 int btf__add_func_param(struct btf *btf, const char *name, int type_id)
2365 {
2366 	struct btf_type *t;
2367 	struct btf_param *p;
2368 	int sz, name_off = 0;
2369 
2370 	if (validate_type_id(type_id))
2371 		return libbpf_err(-EINVAL);
2372 
2373 	/* last type should be BTF_KIND_FUNC_PROTO */
2374 	if (btf->nr_types == 0)
2375 		return libbpf_err(-EINVAL);
2376 	t = btf_last_type(btf);
2377 	if (!btf_is_func_proto(t))
2378 		return libbpf_err(-EINVAL);
2379 
2380 	/* decompose and invalidate raw data */
2381 	if (btf_ensure_modifiable(btf))
2382 		return libbpf_err(-ENOMEM);
2383 
2384 	sz = sizeof(struct btf_param);
2385 	p = btf_add_type_mem(btf, sz);
2386 	if (!p)
2387 		return libbpf_err(-ENOMEM);
2388 
2389 	if (name && name[0]) {
2390 		name_off = btf__add_str(btf, name);
2391 		if (name_off < 0)
2392 			return name_off;
2393 	}
2394 
2395 	p->name_off = name_off;
2396 	p->type = type_id;
2397 
2398 	/* update parent type's vlen */
2399 	t = btf_last_type(btf);
2400 	btf_type_inc_vlen(t);
2401 
2402 	btf->hdr->type_len += sz;
2403 	btf->hdr->str_off += sz;
2404 	return 0;
2405 }
2406 
2407 /*
2408  * Append new BTF_KIND_VAR type with:
2409  *   - *name* - non-empty/non-NULL name;
2410  *   - *linkage* - variable linkage, one of BTF_VAR_STATIC,
2411  *     BTF_VAR_GLOBAL_ALLOCATED, or BTF_VAR_GLOBAL_EXTERN;
2412  *   - *type_id* - type ID of the type describing the type of the variable.
2413  * Returns:
2414  *   - >0, type ID of newly added BTF type;
2415  *   - <0, on error.
2416  */
2417 int btf__add_var(struct btf *btf, const char *name, int linkage, int type_id)
2418 {
2419 	struct btf_type *t;
2420 	struct btf_var *v;
2421 	int sz, name_off;
2422 
2423 	/* non-empty name */
2424 	if (!name || !name[0])
2425 		return libbpf_err(-EINVAL);
2426 	if (linkage != BTF_VAR_STATIC && linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2427 	    linkage != BTF_VAR_GLOBAL_EXTERN)
2428 		return libbpf_err(-EINVAL);
2429 	if (validate_type_id(type_id))
2430 		return libbpf_err(-EINVAL);
2431 
2432 	/* deconstruct BTF, if necessary, and invalidate raw_data */
2433 	if (btf_ensure_modifiable(btf))
2434 		return libbpf_err(-ENOMEM);
2435 
2436 	sz = sizeof(struct btf_type) + sizeof(struct btf_var);
2437 	t = btf_add_type_mem(btf, sz);
2438 	if (!t)
2439 		return libbpf_err(-ENOMEM);
2440 
2441 	name_off = btf__add_str(btf, name);
2442 	if (name_off < 0)
2443 		return name_off;
2444 
2445 	t->name_off = name_off;
2446 	t->info = btf_type_info(BTF_KIND_VAR, 0, 0);
2447 	t->type = type_id;
2448 
2449 	v = btf_var(t);
2450 	v->linkage = linkage;
2451 
2452 	return btf_commit_type(btf, sz);
2453 }
2454 
2455 /*
2456  * Append new BTF_KIND_DATASEC type with:
2457  *   - *name* - non-empty/non-NULL name;
2458  *   - *byte_sz* - data section size, in bytes.
2459  *
2460  * Data section is initially empty. Variables info can be added with
2461  * btf__add_datasec_var_info() calls, after btf__add_datasec() succeeds.
2462  *
2463  * Returns:
2464  *   - >0, type ID of newly added BTF type;
2465  *   - <0, on error.
2466  */
2467 int btf__add_datasec(struct btf *btf, const char *name, __u32 byte_sz)
2468 {
2469 	struct btf_type *t;
2470 	int sz, name_off;
2471 
2472 	/* non-empty name */
2473 	if (!name || !name[0])
2474 		return libbpf_err(-EINVAL);
2475 
2476 	if (btf_ensure_modifiable(btf))
2477 		return libbpf_err(-ENOMEM);
2478 
2479 	sz = sizeof(struct btf_type);
2480 	t = btf_add_type_mem(btf, sz);
2481 	if (!t)
2482 		return libbpf_err(-ENOMEM);
2483 
2484 	name_off = btf__add_str(btf, name);
2485 	if (name_off < 0)
2486 		return name_off;
2487 
2488 	/* start with vlen=0, which will be update as var_secinfos are added */
2489 	t->name_off = name_off;
2490 	t->info = btf_type_info(BTF_KIND_DATASEC, 0, 0);
2491 	t->size = byte_sz;
2492 
2493 	return btf_commit_type(btf, sz);
2494 }
2495 
2496 /*
2497  * Append new data section variable information entry for current DATASEC type:
2498  *   - *var_type_id* - type ID, describing type of the variable;
2499  *   - *offset* - variable offset within data section, in bytes;
2500  *   - *byte_sz* - variable size, in bytes.
2501  *
2502  * Returns:
2503  *   -  0, on success;
2504  *   - <0, on error.
2505  */
2506 int btf__add_datasec_var_info(struct btf *btf, int var_type_id, __u32 offset, __u32 byte_sz)
2507 {
2508 	struct btf_type *t;
2509 	struct btf_var_secinfo *v;
2510 	int sz;
2511 
2512 	/* last type should be BTF_KIND_DATASEC */
2513 	if (btf->nr_types == 0)
2514 		return libbpf_err(-EINVAL);
2515 	t = btf_last_type(btf);
2516 	if (!btf_is_datasec(t))
2517 		return libbpf_err(-EINVAL);
2518 
2519 	if (validate_type_id(var_type_id))
2520 		return libbpf_err(-EINVAL);
2521 
2522 	/* decompose and invalidate raw data */
2523 	if (btf_ensure_modifiable(btf))
2524 		return libbpf_err(-ENOMEM);
2525 
2526 	sz = sizeof(struct btf_var_secinfo);
2527 	v = btf_add_type_mem(btf, sz);
2528 	if (!v)
2529 		return libbpf_err(-ENOMEM);
2530 
2531 	v->type = var_type_id;
2532 	v->offset = offset;
2533 	v->size = byte_sz;
2534 
2535 	/* update parent type's vlen */
2536 	t = btf_last_type(btf);
2537 	btf_type_inc_vlen(t);
2538 
2539 	btf->hdr->type_len += sz;
2540 	btf->hdr->str_off += sz;
2541 	return 0;
2542 }
2543 
2544 /*
2545  * Append new BTF_KIND_DECL_TAG type with:
2546  *   - *value* - non-empty/non-NULL string;
2547  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2548  *   - *component_idx* - -1 for tagging reference type, otherwise struct/union
2549  *     member or function argument index;
2550  * Returns:
2551  *   - >0, type ID of newly added BTF type;
2552  *   - <0, on error.
2553  */
2554 int btf__add_decl_tag(struct btf *btf, const char *value, int ref_type_id,
2555 		 int component_idx)
2556 {
2557 	struct btf_type *t;
2558 	int sz, value_off;
2559 
2560 	if (!value || !value[0] || component_idx < -1)
2561 		return libbpf_err(-EINVAL);
2562 
2563 	if (validate_type_id(ref_type_id))
2564 		return libbpf_err(-EINVAL);
2565 
2566 	if (btf_ensure_modifiable(btf))
2567 		return libbpf_err(-ENOMEM);
2568 
2569 	sz = sizeof(struct btf_type) + sizeof(struct btf_decl_tag);
2570 	t = btf_add_type_mem(btf, sz);
2571 	if (!t)
2572 		return libbpf_err(-ENOMEM);
2573 
2574 	value_off = btf__add_str(btf, value);
2575 	if (value_off < 0)
2576 		return value_off;
2577 
2578 	t->name_off = value_off;
2579 	t->info = btf_type_info(BTF_KIND_DECL_TAG, 0, false);
2580 	t->type = ref_type_id;
2581 	btf_decl_tag(t)->component_idx = component_idx;
2582 
2583 	return btf_commit_type(btf, sz);
2584 }
2585 
2586 struct btf_ext_sec_setup_param {
2587 	__u32 off;
2588 	__u32 len;
2589 	__u32 min_rec_size;
2590 	struct btf_ext_info *ext_info;
2591 	const char *desc;
2592 };
2593 
2594 static int btf_ext_setup_info(struct btf_ext *btf_ext,
2595 			      struct btf_ext_sec_setup_param *ext_sec)
2596 {
2597 	const struct btf_ext_info_sec *sinfo;
2598 	struct btf_ext_info *ext_info;
2599 	__u32 info_left, record_size;
2600 	/* The start of the info sec (including the __u32 record_size). */
2601 	void *info;
2602 
2603 	if (ext_sec->len == 0)
2604 		return 0;
2605 
2606 	if (ext_sec->off & 0x03) {
2607 		pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
2608 		     ext_sec->desc);
2609 		return -EINVAL;
2610 	}
2611 
2612 	info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
2613 	info_left = ext_sec->len;
2614 
2615 	if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
2616 		pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
2617 			 ext_sec->desc, ext_sec->off, ext_sec->len);
2618 		return -EINVAL;
2619 	}
2620 
2621 	/* At least a record size */
2622 	if (info_left < sizeof(__u32)) {
2623 		pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
2624 		return -EINVAL;
2625 	}
2626 
2627 	/* The record size needs to meet the minimum standard */
2628 	record_size = *(__u32 *)info;
2629 	if (record_size < ext_sec->min_rec_size ||
2630 	    record_size & 0x03) {
2631 		pr_debug("%s section in .BTF.ext has invalid record size %u\n",
2632 			 ext_sec->desc, record_size);
2633 		return -EINVAL;
2634 	}
2635 
2636 	sinfo = info + sizeof(__u32);
2637 	info_left -= sizeof(__u32);
2638 
2639 	/* If no records, return failure now so .BTF.ext won't be used. */
2640 	if (!info_left) {
2641 		pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
2642 		return -EINVAL;
2643 	}
2644 
2645 	while (info_left) {
2646 		unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
2647 		__u64 total_record_size;
2648 		__u32 num_records;
2649 
2650 		if (info_left < sec_hdrlen) {
2651 			pr_debug("%s section header is not found in .BTF.ext\n",
2652 			     ext_sec->desc);
2653 			return -EINVAL;
2654 		}
2655 
2656 		num_records = sinfo->num_info;
2657 		if (num_records == 0) {
2658 			pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2659 			     ext_sec->desc);
2660 			return -EINVAL;
2661 		}
2662 
2663 		total_record_size = sec_hdrlen +
2664 				    (__u64)num_records * record_size;
2665 		if (info_left < total_record_size) {
2666 			pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2667 			     ext_sec->desc);
2668 			return -EINVAL;
2669 		}
2670 
2671 		info_left -= total_record_size;
2672 		sinfo = (void *)sinfo + total_record_size;
2673 	}
2674 
2675 	ext_info = ext_sec->ext_info;
2676 	ext_info->len = ext_sec->len - sizeof(__u32);
2677 	ext_info->rec_size = record_size;
2678 	ext_info->info = info + sizeof(__u32);
2679 
2680 	return 0;
2681 }
2682 
2683 static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
2684 {
2685 	struct btf_ext_sec_setup_param param = {
2686 		.off = btf_ext->hdr->func_info_off,
2687 		.len = btf_ext->hdr->func_info_len,
2688 		.min_rec_size = sizeof(struct bpf_func_info_min),
2689 		.ext_info = &btf_ext->func_info,
2690 		.desc = "func_info"
2691 	};
2692 
2693 	return btf_ext_setup_info(btf_ext, &param);
2694 }
2695 
2696 static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
2697 {
2698 	struct btf_ext_sec_setup_param param = {
2699 		.off = btf_ext->hdr->line_info_off,
2700 		.len = btf_ext->hdr->line_info_len,
2701 		.min_rec_size = sizeof(struct bpf_line_info_min),
2702 		.ext_info = &btf_ext->line_info,
2703 		.desc = "line_info",
2704 	};
2705 
2706 	return btf_ext_setup_info(btf_ext, &param);
2707 }
2708 
2709 static int btf_ext_setup_core_relos(struct btf_ext *btf_ext)
2710 {
2711 	struct btf_ext_sec_setup_param param = {
2712 		.off = btf_ext->hdr->core_relo_off,
2713 		.len = btf_ext->hdr->core_relo_len,
2714 		.min_rec_size = sizeof(struct bpf_core_relo),
2715 		.ext_info = &btf_ext->core_relo_info,
2716 		.desc = "core_relo",
2717 	};
2718 
2719 	return btf_ext_setup_info(btf_ext, &param);
2720 }
2721 
2722 static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
2723 {
2724 	const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
2725 
2726 	if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
2727 	    data_size < hdr->hdr_len) {
2728 		pr_debug("BTF.ext header not found");
2729 		return -EINVAL;
2730 	}
2731 
2732 	if (hdr->magic == bswap_16(BTF_MAGIC)) {
2733 		pr_warn("BTF.ext in non-native endianness is not supported\n");
2734 		return -ENOTSUP;
2735 	} else if (hdr->magic != BTF_MAGIC) {
2736 		pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
2737 		return -EINVAL;
2738 	}
2739 
2740 	if (hdr->version != BTF_VERSION) {
2741 		pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
2742 		return -ENOTSUP;
2743 	}
2744 
2745 	if (hdr->flags) {
2746 		pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
2747 		return -ENOTSUP;
2748 	}
2749 
2750 	if (data_size == hdr->hdr_len) {
2751 		pr_debug("BTF.ext has no data\n");
2752 		return -EINVAL;
2753 	}
2754 
2755 	return 0;
2756 }
2757 
2758 void btf_ext__free(struct btf_ext *btf_ext)
2759 {
2760 	if (IS_ERR_OR_NULL(btf_ext))
2761 		return;
2762 	free(btf_ext->data);
2763 	free(btf_ext);
2764 }
2765 
2766 struct btf_ext *btf_ext__new(const __u8 *data, __u32 size)
2767 {
2768 	struct btf_ext *btf_ext;
2769 	int err;
2770 
2771 	btf_ext = calloc(1, sizeof(struct btf_ext));
2772 	if (!btf_ext)
2773 		return libbpf_err_ptr(-ENOMEM);
2774 
2775 	btf_ext->data_size = size;
2776 	btf_ext->data = malloc(size);
2777 	if (!btf_ext->data) {
2778 		err = -ENOMEM;
2779 		goto done;
2780 	}
2781 	memcpy(btf_ext->data, data, size);
2782 
2783 	err = btf_ext_parse_hdr(btf_ext->data, size);
2784 	if (err)
2785 		goto done;
2786 
2787 	if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, line_info_len)) {
2788 		err = -EINVAL;
2789 		goto done;
2790 	}
2791 
2792 	err = btf_ext_setup_func_info(btf_ext);
2793 	if (err)
2794 		goto done;
2795 
2796 	err = btf_ext_setup_line_info(btf_ext);
2797 	if (err)
2798 		goto done;
2799 
2800 	if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, core_relo_len)) {
2801 		err = -EINVAL;
2802 		goto done;
2803 	}
2804 
2805 	err = btf_ext_setup_core_relos(btf_ext);
2806 	if (err)
2807 		goto done;
2808 
2809 done:
2810 	if (err) {
2811 		btf_ext__free(btf_ext);
2812 		return libbpf_err_ptr(err);
2813 	}
2814 
2815 	return btf_ext;
2816 }
2817 
2818 const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size)
2819 {
2820 	*size = btf_ext->data_size;
2821 	return btf_ext->data;
2822 }
2823 
2824 static int btf_ext_reloc_info(const struct btf *btf,
2825 			      const struct btf_ext_info *ext_info,
2826 			      const char *sec_name, __u32 insns_cnt,
2827 			      void **info, __u32 *cnt)
2828 {
2829 	__u32 sec_hdrlen = sizeof(struct btf_ext_info_sec);
2830 	__u32 i, record_size, existing_len, records_len;
2831 	struct btf_ext_info_sec *sinfo;
2832 	const char *info_sec_name;
2833 	__u64 remain_len;
2834 	void *data;
2835 
2836 	record_size = ext_info->rec_size;
2837 	sinfo = ext_info->info;
2838 	remain_len = ext_info->len;
2839 	while (remain_len > 0) {
2840 		records_len = sinfo->num_info * record_size;
2841 		info_sec_name = btf__name_by_offset(btf, sinfo->sec_name_off);
2842 		if (strcmp(info_sec_name, sec_name)) {
2843 			remain_len -= sec_hdrlen + records_len;
2844 			sinfo = (void *)sinfo + sec_hdrlen + records_len;
2845 			continue;
2846 		}
2847 
2848 		existing_len = (*cnt) * record_size;
2849 		data = realloc(*info, existing_len + records_len);
2850 		if (!data)
2851 			return libbpf_err(-ENOMEM);
2852 
2853 		memcpy(data + existing_len, sinfo->data, records_len);
2854 		/* adjust insn_off only, the rest data will be passed
2855 		 * to the kernel.
2856 		 */
2857 		for (i = 0; i < sinfo->num_info; i++) {
2858 			__u32 *insn_off;
2859 
2860 			insn_off = data + existing_len + (i * record_size);
2861 			*insn_off = *insn_off / sizeof(struct bpf_insn) + insns_cnt;
2862 		}
2863 		*info = data;
2864 		*cnt += sinfo->num_info;
2865 		return 0;
2866 	}
2867 
2868 	return libbpf_err(-ENOENT);
2869 }
2870 
2871 int btf_ext__reloc_func_info(const struct btf *btf,
2872 			     const struct btf_ext *btf_ext,
2873 			     const char *sec_name, __u32 insns_cnt,
2874 			     void **func_info, __u32 *cnt)
2875 {
2876 	return btf_ext_reloc_info(btf, &btf_ext->func_info, sec_name,
2877 				  insns_cnt, func_info, cnt);
2878 }
2879 
2880 int btf_ext__reloc_line_info(const struct btf *btf,
2881 			     const struct btf_ext *btf_ext,
2882 			     const char *sec_name, __u32 insns_cnt,
2883 			     void **line_info, __u32 *cnt)
2884 {
2885 	return btf_ext_reloc_info(btf, &btf_ext->line_info, sec_name,
2886 				  insns_cnt, line_info, cnt);
2887 }
2888 
2889 __u32 btf_ext__func_info_rec_size(const struct btf_ext *btf_ext)
2890 {
2891 	return btf_ext->func_info.rec_size;
2892 }
2893 
2894 __u32 btf_ext__line_info_rec_size(const struct btf_ext *btf_ext)
2895 {
2896 	return btf_ext->line_info.rec_size;
2897 }
2898 
2899 struct btf_dedup;
2900 
2901 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts);
2902 static void btf_dedup_free(struct btf_dedup *d);
2903 static int btf_dedup_prep(struct btf_dedup *d);
2904 static int btf_dedup_strings(struct btf_dedup *d);
2905 static int btf_dedup_prim_types(struct btf_dedup *d);
2906 static int btf_dedup_struct_types(struct btf_dedup *d);
2907 static int btf_dedup_ref_types(struct btf_dedup *d);
2908 static int btf_dedup_compact_types(struct btf_dedup *d);
2909 static int btf_dedup_remap_types(struct btf_dedup *d);
2910 
2911 /*
2912  * Deduplicate BTF types and strings.
2913  *
2914  * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
2915  * section with all BTF type descriptors and string data. It overwrites that
2916  * memory in-place with deduplicated types and strings without any loss of
2917  * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
2918  * is provided, all the strings referenced from .BTF.ext section are honored
2919  * and updated to point to the right offsets after deduplication.
2920  *
2921  * If function returns with error, type/string data might be garbled and should
2922  * be discarded.
2923  *
2924  * More verbose and detailed description of both problem btf_dedup is solving,
2925  * as well as solution could be found at:
2926  * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
2927  *
2928  * Problem description and justification
2929  * =====================================
2930  *
2931  * BTF type information is typically emitted either as a result of conversion
2932  * from DWARF to BTF or directly by compiler. In both cases, each compilation
2933  * unit contains information about a subset of all the types that are used
2934  * in an application. These subsets are frequently overlapping and contain a lot
2935  * of duplicated information when later concatenated together into a single
2936  * binary. This algorithm ensures that each unique type is represented by single
2937  * BTF type descriptor, greatly reducing resulting size of BTF data.
2938  *
2939  * Compilation unit isolation and subsequent duplication of data is not the only
2940  * problem. The same type hierarchy (e.g., struct and all the type that struct
2941  * references) in different compilation units can be represented in BTF to
2942  * various degrees of completeness (or, rather, incompleteness) due to
2943  * struct/union forward declarations.
2944  *
2945  * Let's take a look at an example, that we'll use to better understand the
2946  * problem (and solution). Suppose we have two compilation units, each using
2947  * same `struct S`, but each of them having incomplete type information about
2948  * struct's fields:
2949  *
2950  * // CU #1:
2951  * struct S;
2952  * struct A {
2953  *	int a;
2954  *	struct A* self;
2955  *	struct S* parent;
2956  * };
2957  * struct B;
2958  * struct S {
2959  *	struct A* a_ptr;
2960  *	struct B* b_ptr;
2961  * };
2962  *
2963  * // CU #2:
2964  * struct S;
2965  * struct A;
2966  * struct B {
2967  *	int b;
2968  *	struct B* self;
2969  *	struct S* parent;
2970  * };
2971  * struct S {
2972  *	struct A* a_ptr;
2973  *	struct B* b_ptr;
2974  * };
2975  *
2976  * In case of CU #1, BTF data will know only that `struct B` exist (but no
2977  * more), but will know the complete type information about `struct A`. While
2978  * for CU #2, it will know full type information about `struct B`, but will
2979  * only know about forward declaration of `struct A` (in BTF terms, it will
2980  * have `BTF_KIND_FWD` type descriptor with name `B`).
2981  *
2982  * This compilation unit isolation means that it's possible that there is no
2983  * single CU with complete type information describing structs `S`, `A`, and
2984  * `B`. Also, we might get tons of duplicated and redundant type information.
2985  *
2986  * Additional complication we need to keep in mind comes from the fact that
2987  * types, in general, can form graphs containing cycles, not just DAGs.
2988  *
2989  * While algorithm does deduplication, it also merges and resolves type
2990  * information (unless disabled throught `struct btf_opts`), whenever possible.
2991  * E.g., in the example above with two compilation units having partial type
2992  * information for structs `A` and `B`, the output of algorithm will emit
2993  * a single copy of each BTF type that describes structs `A`, `B`, and `S`
2994  * (as well as type information for `int` and pointers), as if they were defined
2995  * in a single compilation unit as:
2996  *
2997  * struct A {
2998  *	int a;
2999  *	struct A* self;
3000  *	struct S* parent;
3001  * };
3002  * struct B {
3003  *	int b;
3004  *	struct B* self;
3005  *	struct S* parent;
3006  * };
3007  * struct S {
3008  *	struct A* a_ptr;
3009  *	struct B* b_ptr;
3010  * };
3011  *
3012  * Algorithm summary
3013  * =================
3014  *
3015  * Algorithm completes its work in 6 separate passes:
3016  *
3017  * 1. Strings deduplication.
3018  * 2. Primitive types deduplication (int, enum, fwd).
3019  * 3. Struct/union types deduplication.
3020  * 4. Reference types deduplication (pointers, typedefs, arrays, funcs, func
3021  *    protos, and const/volatile/restrict modifiers).
3022  * 5. Types compaction.
3023  * 6. Types remapping.
3024  *
3025  * Algorithm determines canonical type descriptor, which is a single
3026  * representative type for each truly unique type. This canonical type is the
3027  * one that will go into final deduplicated BTF type information. For
3028  * struct/unions, it is also the type that algorithm will merge additional type
3029  * information into (while resolving FWDs), as it discovers it from data in
3030  * other CUs. Each input BTF type eventually gets either mapped to itself, if
3031  * that type is canonical, or to some other type, if that type is equivalent
3032  * and was chosen as canonical representative. This mapping is stored in
3033  * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
3034  * FWD type got resolved to.
3035  *
3036  * To facilitate fast discovery of canonical types, we also maintain canonical
3037  * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
3038  * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
3039  * that match that signature. With sufficiently good choice of type signature
3040  * hashing function, we can limit number of canonical types for each unique type
3041  * signature to a very small number, allowing to find canonical type for any
3042  * duplicated type very quickly.
3043  *
3044  * Struct/union deduplication is the most critical part and algorithm for
3045  * deduplicating structs/unions is described in greater details in comments for
3046  * `btf_dedup_is_equiv` function.
3047  */
3048 
3049 DEFAULT_VERSION(btf__dedup_v0_6_0, btf__dedup, LIBBPF_0.6.0)
3050 int btf__dedup_v0_6_0(struct btf *btf, const struct btf_dedup_opts *opts)
3051 {
3052 	struct btf_dedup *d;
3053 	int err;
3054 
3055 	if (!OPTS_VALID(opts, btf_dedup_opts))
3056 		return libbpf_err(-EINVAL);
3057 
3058 	d = btf_dedup_new(btf, opts);
3059 	if (IS_ERR(d)) {
3060 		pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
3061 		return libbpf_err(-EINVAL);
3062 	}
3063 
3064 	if (btf_ensure_modifiable(btf)) {
3065 		err = -ENOMEM;
3066 		goto done;
3067 	}
3068 
3069 	err = btf_dedup_prep(d);
3070 	if (err) {
3071 		pr_debug("btf_dedup_prep failed:%d\n", err);
3072 		goto done;
3073 	}
3074 	err = btf_dedup_strings(d);
3075 	if (err < 0) {
3076 		pr_debug("btf_dedup_strings failed:%d\n", err);
3077 		goto done;
3078 	}
3079 	err = btf_dedup_prim_types(d);
3080 	if (err < 0) {
3081 		pr_debug("btf_dedup_prim_types failed:%d\n", err);
3082 		goto done;
3083 	}
3084 	err = btf_dedup_struct_types(d);
3085 	if (err < 0) {
3086 		pr_debug("btf_dedup_struct_types failed:%d\n", err);
3087 		goto done;
3088 	}
3089 	err = btf_dedup_ref_types(d);
3090 	if (err < 0) {
3091 		pr_debug("btf_dedup_ref_types failed:%d\n", err);
3092 		goto done;
3093 	}
3094 	err = btf_dedup_compact_types(d);
3095 	if (err < 0) {
3096 		pr_debug("btf_dedup_compact_types failed:%d\n", err);
3097 		goto done;
3098 	}
3099 	err = btf_dedup_remap_types(d);
3100 	if (err < 0) {
3101 		pr_debug("btf_dedup_remap_types failed:%d\n", err);
3102 		goto done;
3103 	}
3104 
3105 done:
3106 	btf_dedup_free(d);
3107 	return libbpf_err(err);
3108 }
3109 
3110 COMPAT_VERSION(btf__dedup_deprecated, btf__dedup, LIBBPF_0.0.2)
3111 int btf__dedup_deprecated(struct btf *btf, struct btf_ext *btf_ext, const void *unused_opts)
3112 {
3113 	LIBBPF_OPTS(btf_dedup_opts, opts, .btf_ext = btf_ext);
3114 
3115 	if (unused_opts) {
3116 		pr_warn("please use new version of btf__dedup() that supports options\n");
3117 		return libbpf_err(-ENOTSUP);
3118 	}
3119 
3120 	return btf__dedup(btf, &opts);
3121 }
3122 
3123 #define BTF_UNPROCESSED_ID ((__u32)-1)
3124 #define BTF_IN_PROGRESS_ID ((__u32)-2)
3125 
3126 struct btf_dedup {
3127 	/* .BTF section to be deduped in-place */
3128 	struct btf *btf;
3129 	/*
3130 	 * Optional .BTF.ext section. When provided, any strings referenced
3131 	 * from it will be taken into account when deduping strings
3132 	 */
3133 	struct btf_ext *btf_ext;
3134 	/*
3135 	 * This is a map from any type's signature hash to a list of possible
3136 	 * canonical representative type candidates. Hash collisions are
3137 	 * ignored, so even types of various kinds can share same list of
3138 	 * candidates, which is fine because we rely on subsequent
3139 	 * btf_xxx_equal() checks to authoritatively verify type equality.
3140 	 */
3141 	struct hashmap *dedup_table;
3142 	/* Canonical types map */
3143 	__u32 *map;
3144 	/* Hypothetical mapping, used during type graph equivalence checks */
3145 	__u32 *hypot_map;
3146 	__u32 *hypot_list;
3147 	size_t hypot_cnt;
3148 	size_t hypot_cap;
3149 	/* Whether hypothetical mapping, if successful, would need to adjust
3150 	 * already canonicalized types (due to a new forward declaration to
3151 	 * concrete type resolution). In such case, during split BTF dedup
3152 	 * candidate type would still be considered as different, because base
3153 	 * BTF is considered to be immutable.
3154 	 */
3155 	bool hypot_adjust_canon;
3156 	/* Various option modifying behavior of algorithm */
3157 	struct btf_dedup_opts opts;
3158 	/* temporary strings deduplication state */
3159 	struct strset *strs_set;
3160 };
3161 
3162 static long hash_combine(long h, long value)
3163 {
3164 	return h * 31 + value;
3165 }
3166 
3167 #define for_each_dedup_cand(d, node, hash) \
3168 	hashmap__for_each_key_entry(d->dedup_table, node, (void *)hash)
3169 
3170 static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
3171 {
3172 	return hashmap__append(d->dedup_table,
3173 			       (void *)hash, (void *)(long)type_id);
3174 }
3175 
3176 static int btf_dedup_hypot_map_add(struct btf_dedup *d,
3177 				   __u32 from_id, __u32 to_id)
3178 {
3179 	if (d->hypot_cnt == d->hypot_cap) {
3180 		__u32 *new_list;
3181 
3182 		d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
3183 		new_list = libbpf_reallocarray(d->hypot_list, d->hypot_cap, sizeof(__u32));
3184 		if (!new_list)
3185 			return -ENOMEM;
3186 		d->hypot_list = new_list;
3187 	}
3188 	d->hypot_list[d->hypot_cnt++] = from_id;
3189 	d->hypot_map[from_id] = to_id;
3190 	return 0;
3191 }
3192 
3193 static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
3194 {
3195 	int i;
3196 
3197 	for (i = 0; i < d->hypot_cnt; i++)
3198 		d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
3199 	d->hypot_cnt = 0;
3200 	d->hypot_adjust_canon = false;
3201 }
3202 
3203 static void btf_dedup_free(struct btf_dedup *d)
3204 {
3205 	hashmap__free(d->dedup_table);
3206 	d->dedup_table = NULL;
3207 
3208 	free(d->map);
3209 	d->map = NULL;
3210 
3211 	free(d->hypot_map);
3212 	d->hypot_map = NULL;
3213 
3214 	free(d->hypot_list);
3215 	d->hypot_list = NULL;
3216 
3217 	free(d);
3218 }
3219 
3220 static size_t btf_dedup_identity_hash_fn(const void *key, void *ctx)
3221 {
3222 	return (size_t)key;
3223 }
3224 
3225 static size_t btf_dedup_collision_hash_fn(const void *key, void *ctx)
3226 {
3227 	return 0;
3228 }
3229 
3230 static bool btf_dedup_equal_fn(const void *k1, const void *k2, void *ctx)
3231 {
3232 	return k1 == k2;
3233 }
3234 
3235 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts)
3236 {
3237 	struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
3238 	hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
3239 	int i, err = 0, type_cnt;
3240 
3241 	if (!d)
3242 		return ERR_PTR(-ENOMEM);
3243 
3244 	if (OPTS_GET(opts, force_collisions, false))
3245 		hash_fn = btf_dedup_collision_hash_fn;
3246 
3247 	d->btf = btf;
3248 	d->btf_ext = OPTS_GET(opts, btf_ext, NULL);
3249 
3250 	d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
3251 	if (IS_ERR(d->dedup_table)) {
3252 		err = PTR_ERR(d->dedup_table);
3253 		d->dedup_table = NULL;
3254 		goto done;
3255 	}
3256 
3257 	type_cnt = btf__type_cnt(btf);
3258 	d->map = malloc(sizeof(__u32) * type_cnt);
3259 	if (!d->map) {
3260 		err = -ENOMEM;
3261 		goto done;
3262 	}
3263 	/* special BTF "void" type is made canonical immediately */
3264 	d->map[0] = 0;
3265 	for (i = 1; i < type_cnt; i++) {
3266 		struct btf_type *t = btf_type_by_id(d->btf, i);
3267 
3268 		/* VAR and DATASEC are never deduped and are self-canonical */
3269 		if (btf_is_var(t) || btf_is_datasec(t))
3270 			d->map[i] = i;
3271 		else
3272 			d->map[i] = BTF_UNPROCESSED_ID;
3273 	}
3274 
3275 	d->hypot_map = malloc(sizeof(__u32) * type_cnt);
3276 	if (!d->hypot_map) {
3277 		err = -ENOMEM;
3278 		goto done;
3279 	}
3280 	for (i = 0; i < type_cnt; i++)
3281 		d->hypot_map[i] = BTF_UNPROCESSED_ID;
3282 
3283 done:
3284 	if (err) {
3285 		btf_dedup_free(d);
3286 		return ERR_PTR(err);
3287 	}
3288 
3289 	return d;
3290 }
3291 
3292 /*
3293  * Iterate over all possible places in .BTF and .BTF.ext that can reference
3294  * string and pass pointer to it to a provided callback `fn`.
3295  */
3296 static int btf_for_each_str_off(struct btf_dedup *d, str_off_visit_fn fn, void *ctx)
3297 {
3298 	int i, r;
3299 
3300 	for (i = 0; i < d->btf->nr_types; i++) {
3301 		struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
3302 
3303 		r = btf_type_visit_str_offs(t, fn, ctx);
3304 		if (r)
3305 			return r;
3306 	}
3307 
3308 	if (!d->btf_ext)
3309 		return 0;
3310 
3311 	r = btf_ext_visit_str_offs(d->btf_ext, fn, ctx);
3312 	if (r)
3313 		return r;
3314 
3315 	return 0;
3316 }
3317 
3318 static int strs_dedup_remap_str_off(__u32 *str_off_ptr, void *ctx)
3319 {
3320 	struct btf_dedup *d = ctx;
3321 	__u32 str_off = *str_off_ptr;
3322 	const char *s;
3323 	int off, err;
3324 
3325 	/* don't touch empty string or string in main BTF */
3326 	if (str_off == 0 || str_off < d->btf->start_str_off)
3327 		return 0;
3328 
3329 	s = btf__str_by_offset(d->btf, str_off);
3330 	if (d->btf->base_btf) {
3331 		err = btf__find_str(d->btf->base_btf, s);
3332 		if (err >= 0) {
3333 			*str_off_ptr = err;
3334 			return 0;
3335 		}
3336 		if (err != -ENOENT)
3337 			return err;
3338 	}
3339 
3340 	off = strset__add_str(d->strs_set, s);
3341 	if (off < 0)
3342 		return off;
3343 
3344 	*str_off_ptr = d->btf->start_str_off + off;
3345 	return 0;
3346 }
3347 
3348 /*
3349  * Dedup string and filter out those that are not referenced from either .BTF
3350  * or .BTF.ext (if provided) sections.
3351  *
3352  * This is done by building index of all strings in BTF's string section,
3353  * then iterating over all entities that can reference strings (e.g., type
3354  * names, struct field names, .BTF.ext line info, etc) and marking corresponding
3355  * strings as used. After that all used strings are deduped and compacted into
3356  * sequential blob of memory and new offsets are calculated. Then all the string
3357  * references are iterated again and rewritten using new offsets.
3358  */
3359 static int btf_dedup_strings(struct btf_dedup *d)
3360 {
3361 	int err;
3362 
3363 	if (d->btf->strs_deduped)
3364 		return 0;
3365 
3366 	d->strs_set = strset__new(BTF_MAX_STR_OFFSET, NULL, 0);
3367 	if (IS_ERR(d->strs_set)) {
3368 		err = PTR_ERR(d->strs_set);
3369 		goto err_out;
3370 	}
3371 
3372 	if (!d->btf->base_btf) {
3373 		/* insert empty string; we won't be looking it up during strings
3374 		 * dedup, but it's good to have it for generic BTF string lookups
3375 		 */
3376 		err = strset__add_str(d->strs_set, "");
3377 		if (err < 0)
3378 			goto err_out;
3379 	}
3380 
3381 	/* remap string offsets */
3382 	err = btf_for_each_str_off(d, strs_dedup_remap_str_off, d);
3383 	if (err)
3384 		goto err_out;
3385 
3386 	/* replace BTF string data and hash with deduped ones */
3387 	strset__free(d->btf->strs_set);
3388 	d->btf->hdr->str_len = strset__data_size(d->strs_set);
3389 	d->btf->strs_set = d->strs_set;
3390 	d->strs_set = NULL;
3391 	d->btf->strs_deduped = true;
3392 	return 0;
3393 
3394 err_out:
3395 	strset__free(d->strs_set);
3396 	d->strs_set = NULL;
3397 
3398 	return err;
3399 }
3400 
3401 static long btf_hash_common(struct btf_type *t)
3402 {
3403 	long h;
3404 
3405 	h = hash_combine(0, t->name_off);
3406 	h = hash_combine(h, t->info);
3407 	h = hash_combine(h, t->size);
3408 	return h;
3409 }
3410 
3411 static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
3412 {
3413 	return t1->name_off == t2->name_off &&
3414 	       t1->info == t2->info &&
3415 	       t1->size == t2->size;
3416 }
3417 
3418 /* Calculate type signature hash of INT or TAG. */
3419 static long btf_hash_int_decl_tag(struct btf_type *t)
3420 {
3421 	__u32 info = *(__u32 *)(t + 1);
3422 	long h;
3423 
3424 	h = btf_hash_common(t);
3425 	h = hash_combine(h, info);
3426 	return h;
3427 }
3428 
3429 /* Check structural equality of two INTs or TAGs. */
3430 static bool btf_equal_int_tag(struct btf_type *t1, struct btf_type *t2)
3431 {
3432 	__u32 info1, info2;
3433 
3434 	if (!btf_equal_common(t1, t2))
3435 		return false;
3436 	info1 = *(__u32 *)(t1 + 1);
3437 	info2 = *(__u32 *)(t2 + 1);
3438 	return info1 == info2;
3439 }
3440 
3441 /* Calculate type signature hash of ENUM. */
3442 static long btf_hash_enum(struct btf_type *t)
3443 {
3444 	long h;
3445 
3446 	/* don't hash vlen and enum members to support enum fwd resolving */
3447 	h = hash_combine(0, t->name_off);
3448 	h = hash_combine(h, t->info & ~0xffff);
3449 	h = hash_combine(h, t->size);
3450 	return h;
3451 }
3452 
3453 /* Check structural equality of two ENUMs. */
3454 static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
3455 {
3456 	const struct btf_enum *m1, *m2;
3457 	__u16 vlen;
3458 	int i;
3459 
3460 	if (!btf_equal_common(t1, t2))
3461 		return false;
3462 
3463 	vlen = btf_vlen(t1);
3464 	m1 = btf_enum(t1);
3465 	m2 = btf_enum(t2);
3466 	for (i = 0; i < vlen; i++) {
3467 		if (m1->name_off != m2->name_off || m1->val != m2->val)
3468 			return false;
3469 		m1++;
3470 		m2++;
3471 	}
3472 	return true;
3473 }
3474 
3475 static inline bool btf_is_enum_fwd(struct btf_type *t)
3476 {
3477 	return btf_is_enum(t) && btf_vlen(t) == 0;
3478 }
3479 
3480 static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
3481 {
3482 	if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3483 		return btf_equal_enum(t1, t2);
3484 	/* ignore vlen when comparing */
3485 	return t1->name_off == t2->name_off &&
3486 	       (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
3487 	       t1->size == t2->size;
3488 }
3489 
3490 /*
3491  * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
3492  * as referenced type IDs equivalence is established separately during type
3493  * graph equivalence check algorithm.
3494  */
3495 static long btf_hash_struct(struct btf_type *t)
3496 {
3497 	const struct btf_member *member = btf_members(t);
3498 	__u32 vlen = btf_vlen(t);
3499 	long h = btf_hash_common(t);
3500 	int i;
3501 
3502 	for (i = 0; i < vlen; i++) {
3503 		h = hash_combine(h, member->name_off);
3504 		h = hash_combine(h, member->offset);
3505 		/* no hashing of referenced type ID, it can be unresolved yet */
3506 		member++;
3507 	}
3508 	return h;
3509 }
3510 
3511 /*
3512  * Check structural compatibility of two STRUCTs/UNIONs, ignoring referenced
3513  * type IDs. This check is performed during type graph equivalence check and
3514  * referenced types equivalence is checked separately.
3515  */
3516 static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
3517 {
3518 	const struct btf_member *m1, *m2;
3519 	__u16 vlen;
3520 	int i;
3521 
3522 	if (!btf_equal_common(t1, t2))
3523 		return false;
3524 
3525 	vlen = btf_vlen(t1);
3526 	m1 = btf_members(t1);
3527 	m2 = btf_members(t2);
3528 	for (i = 0; i < vlen; i++) {
3529 		if (m1->name_off != m2->name_off || m1->offset != m2->offset)
3530 			return false;
3531 		m1++;
3532 		m2++;
3533 	}
3534 	return true;
3535 }
3536 
3537 /*
3538  * Calculate type signature hash of ARRAY, including referenced type IDs,
3539  * under assumption that they were already resolved to canonical type IDs and
3540  * are not going to change.
3541  */
3542 static long btf_hash_array(struct btf_type *t)
3543 {
3544 	const struct btf_array *info = btf_array(t);
3545 	long h = btf_hash_common(t);
3546 
3547 	h = hash_combine(h, info->type);
3548 	h = hash_combine(h, info->index_type);
3549 	h = hash_combine(h, info->nelems);
3550 	return h;
3551 }
3552 
3553 /*
3554  * Check exact equality of two ARRAYs, taking into account referenced
3555  * type IDs, under assumption that they were already resolved to canonical
3556  * type IDs and are not going to change.
3557  * This function is called during reference types deduplication to compare
3558  * ARRAY to potential canonical representative.
3559  */
3560 static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
3561 {
3562 	const struct btf_array *info1, *info2;
3563 
3564 	if (!btf_equal_common(t1, t2))
3565 		return false;
3566 
3567 	info1 = btf_array(t1);
3568 	info2 = btf_array(t2);
3569 	return info1->type == info2->type &&
3570 	       info1->index_type == info2->index_type &&
3571 	       info1->nelems == info2->nelems;
3572 }
3573 
3574 /*
3575  * Check structural compatibility of two ARRAYs, ignoring referenced type
3576  * IDs. This check is performed during type graph equivalence check and
3577  * referenced types equivalence is checked separately.
3578  */
3579 static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
3580 {
3581 	if (!btf_equal_common(t1, t2))
3582 		return false;
3583 
3584 	return btf_array(t1)->nelems == btf_array(t2)->nelems;
3585 }
3586 
3587 /*
3588  * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
3589  * under assumption that they were already resolved to canonical type IDs and
3590  * are not going to change.
3591  */
3592 static long btf_hash_fnproto(struct btf_type *t)
3593 {
3594 	const struct btf_param *member = btf_params(t);
3595 	__u16 vlen = btf_vlen(t);
3596 	long h = btf_hash_common(t);
3597 	int i;
3598 
3599 	for (i = 0; i < vlen; i++) {
3600 		h = hash_combine(h, member->name_off);
3601 		h = hash_combine(h, member->type);
3602 		member++;
3603 	}
3604 	return h;
3605 }
3606 
3607 /*
3608  * Check exact equality of two FUNC_PROTOs, taking into account referenced
3609  * type IDs, under assumption that they were already resolved to canonical
3610  * type IDs and are not going to change.
3611  * This function is called during reference types deduplication to compare
3612  * FUNC_PROTO to potential canonical representative.
3613  */
3614 static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
3615 {
3616 	const struct btf_param *m1, *m2;
3617 	__u16 vlen;
3618 	int i;
3619 
3620 	if (!btf_equal_common(t1, t2))
3621 		return false;
3622 
3623 	vlen = btf_vlen(t1);
3624 	m1 = btf_params(t1);
3625 	m2 = btf_params(t2);
3626 	for (i = 0; i < vlen; i++) {
3627 		if (m1->name_off != m2->name_off || m1->type != m2->type)
3628 			return false;
3629 		m1++;
3630 		m2++;
3631 	}
3632 	return true;
3633 }
3634 
3635 /*
3636  * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3637  * IDs. This check is performed during type graph equivalence check and
3638  * referenced types equivalence is checked separately.
3639  */
3640 static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
3641 {
3642 	const struct btf_param *m1, *m2;
3643 	__u16 vlen;
3644 	int i;
3645 
3646 	/* skip return type ID */
3647 	if (t1->name_off != t2->name_off || t1->info != t2->info)
3648 		return false;
3649 
3650 	vlen = btf_vlen(t1);
3651 	m1 = btf_params(t1);
3652 	m2 = btf_params(t2);
3653 	for (i = 0; i < vlen; i++) {
3654 		if (m1->name_off != m2->name_off)
3655 			return false;
3656 		m1++;
3657 		m2++;
3658 	}
3659 	return true;
3660 }
3661 
3662 /* Prepare split BTF for deduplication by calculating hashes of base BTF's
3663  * types and initializing the rest of the state (canonical type mapping) for
3664  * the fixed base BTF part.
3665  */
3666 static int btf_dedup_prep(struct btf_dedup *d)
3667 {
3668 	struct btf_type *t;
3669 	int type_id;
3670 	long h;
3671 
3672 	if (!d->btf->base_btf)
3673 		return 0;
3674 
3675 	for (type_id = 1; type_id < d->btf->start_id; type_id++) {
3676 		t = btf_type_by_id(d->btf, type_id);
3677 
3678 		/* all base BTF types are self-canonical by definition */
3679 		d->map[type_id] = type_id;
3680 
3681 		switch (btf_kind(t)) {
3682 		case BTF_KIND_VAR:
3683 		case BTF_KIND_DATASEC:
3684 			/* VAR and DATASEC are never hash/deduplicated */
3685 			continue;
3686 		case BTF_KIND_CONST:
3687 		case BTF_KIND_VOLATILE:
3688 		case BTF_KIND_RESTRICT:
3689 		case BTF_KIND_PTR:
3690 		case BTF_KIND_FWD:
3691 		case BTF_KIND_TYPEDEF:
3692 		case BTF_KIND_FUNC:
3693 		case BTF_KIND_FLOAT:
3694 		case BTF_KIND_TYPE_TAG:
3695 			h = btf_hash_common(t);
3696 			break;
3697 		case BTF_KIND_INT:
3698 		case BTF_KIND_DECL_TAG:
3699 			h = btf_hash_int_decl_tag(t);
3700 			break;
3701 		case BTF_KIND_ENUM:
3702 			h = btf_hash_enum(t);
3703 			break;
3704 		case BTF_KIND_STRUCT:
3705 		case BTF_KIND_UNION:
3706 			h = btf_hash_struct(t);
3707 			break;
3708 		case BTF_KIND_ARRAY:
3709 			h = btf_hash_array(t);
3710 			break;
3711 		case BTF_KIND_FUNC_PROTO:
3712 			h = btf_hash_fnproto(t);
3713 			break;
3714 		default:
3715 			pr_debug("unknown kind %d for type [%d]\n", btf_kind(t), type_id);
3716 			return -EINVAL;
3717 		}
3718 		if (btf_dedup_table_add(d, h, type_id))
3719 			return -ENOMEM;
3720 	}
3721 
3722 	return 0;
3723 }
3724 
3725 /*
3726  * Deduplicate primitive types, that can't reference other types, by calculating
3727  * their type signature hash and comparing them with any possible canonical
3728  * candidate. If no canonical candidate matches, type itself is marked as
3729  * canonical and is added into `btf_dedup->dedup_table` as another candidate.
3730  */
3731 static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
3732 {
3733 	struct btf_type *t = btf_type_by_id(d->btf, type_id);
3734 	struct hashmap_entry *hash_entry;
3735 	struct btf_type *cand;
3736 	/* if we don't find equivalent type, then we are canonical */
3737 	__u32 new_id = type_id;
3738 	__u32 cand_id;
3739 	long h;
3740 
3741 	switch (btf_kind(t)) {
3742 	case BTF_KIND_CONST:
3743 	case BTF_KIND_VOLATILE:
3744 	case BTF_KIND_RESTRICT:
3745 	case BTF_KIND_PTR:
3746 	case BTF_KIND_TYPEDEF:
3747 	case BTF_KIND_ARRAY:
3748 	case BTF_KIND_STRUCT:
3749 	case BTF_KIND_UNION:
3750 	case BTF_KIND_FUNC:
3751 	case BTF_KIND_FUNC_PROTO:
3752 	case BTF_KIND_VAR:
3753 	case BTF_KIND_DATASEC:
3754 	case BTF_KIND_DECL_TAG:
3755 	case BTF_KIND_TYPE_TAG:
3756 		return 0;
3757 
3758 	case BTF_KIND_INT:
3759 		h = btf_hash_int_decl_tag(t);
3760 		for_each_dedup_cand(d, hash_entry, h) {
3761 			cand_id = (__u32)(long)hash_entry->value;
3762 			cand = btf_type_by_id(d->btf, cand_id);
3763 			if (btf_equal_int_tag(t, cand)) {
3764 				new_id = cand_id;
3765 				break;
3766 			}
3767 		}
3768 		break;
3769 
3770 	case BTF_KIND_ENUM:
3771 		h = btf_hash_enum(t);
3772 		for_each_dedup_cand(d, hash_entry, h) {
3773 			cand_id = (__u32)(long)hash_entry->value;
3774 			cand = btf_type_by_id(d->btf, cand_id);
3775 			if (btf_equal_enum(t, cand)) {
3776 				new_id = cand_id;
3777 				break;
3778 			}
3779 			if (btf_compat_enum(t, cand)) {
3780 				if (btf_is_enum_fwd(t)) {
3781 					/* resolve fwd to full enum */
3782 					new_id = cand_id;
3783 					break;
3784 				}
3785 				/* resolve canonical enum fwd to full enum */
3786 				d->map[cand_id] = type_id;
3787 			}
3788 		}
3789 		break;
3790 
3791 	case BTF_KIND_FWD:
3792 	case BTF_KIND_FLOAT:
3793 		h = btf_hash_common(t);
3794 		for_each_dedup_cand(d, hash_entry, h) {
3795 			cand_id = (__u32)(long)hash_entry->value;
3796 			cand = btf_type_by_id(d->btf, cand_id);
3797 			if (btf_equal_common(t, cand)) {
3798 				new_id = cand_id;
3799 				break;
3800 			}
3801 		}
3802 		break;
3803 
3804 	default:
3805 		return -EINVAL;
3806 	}
3807 
3808 	d->map[type_id] = new_id;
3809 	if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
3810 		return -ENOMEM;
3811 
3812 	return 0;
3813 }
3814 
3815 static int btf_dedup_prim_types(struct btf_dedup *d)
3816 {
3817 	int i, err;
3818 
3819 	for (i = 0; i < d->btf->nr_types; i++) {
3820 		err = btf_dedup_prim_type(d, d->btf->start_id + i);
3821 		if (err)
3822 			return err;
3823 	}
3824 	return 0;
3825 }
3826 
3827 /*
3828  * Check whether type is already mapped into canonical one (could be to itself).
3829  */
3830 static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
3831 {
3832 	return d->map[type_id] <= BTF_MAX_NR_TYPES;
3833 }
3834 
3835 /*
3836  * Resolve type ID into its canonical type ID, if any; otherwise return original
3837  * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
3838  * STRUCT/UNION link and resolve it into canonical type ID as well.
3839  */
3840 static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
3841 {
3842 	while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3843 		type_id = d->map[type_id];
3844 	return type_id;
3845 }
3846 
3847 /*
3848  * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
3849  * type ID.
3850  */
3851 static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
3852 {
3853 	__u32 orig_type_id = type_id;
3854 
3855 	if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3856 		return type_id;
3857 
3858 	while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3859 		type_id = d->map[type_id];
3860 
3861 	if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3862 		return type_id;
3863 
3864 	return orig_type_id;
3865 }
3866 
3867 
3868 static inline __u16 btf_fwd_kind(struct btf_type *t)
3869 {
3870 	return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
3871 }
3872 
3873 /* Check if given two types are identical ARRAY definitions */
3874 static int btf_dedup_identical_arrays(struct btf_dedup *d, __u32 id1, __u32 id2)
3875 {
3876 	struct btf_type *t1, *t2;
3877 
3878 	t1 = btf_type_by_id(d->btf, id1);
3879 	t2 = btf_type_by_id(d->btf, id2);
3880 	if (!btf_is_array(t1) || !btf_is_array(t2))
3881 		return 0;
3882 
3883 	return btf_equal_array(t1, t2);
3884 }
3885 
3886 /* Check if given two types are identical STRUCT/UNION definitions */
3887 static bool btf_dedup_identical_structs(struct btf_dedup *d, __u32 id1, __u32 id2)
3888 {
3889 	const struct btf_member *m1, *m2;
3890 	struct btf_type *t1, *t2;
3891 	int n, i;
3892 
3893 	t1 = btf_type_by_id(d->btf, id1);
3894 	t2 = btf_type_by_id(d->btf, id2);
3895 
3896 	if (!btf_is_composite(t1) || btf_kind(t1) != btf_kind(t2))
3897 		return false;
3898 
3899 	if (!btf_shallow_equal_struct(t1, t2))
3900 		return false;
3901 
3902 	m1 = btf_members(t1);
3903 	m2 = btf_members(t2);
3904 	for (i = 0, n = btf_vlen(t1); i < n; i++, m1++, m2++) {
3905 		if (m1->type != m2->type)
3906 			return false;
3907 	}
3908 	return true;
3909 }
3910 
3911 /*
3912  * Check equivalence of BTF type graph formed by candidate struct/union (we'll
3913  * call it "candidate graph" in this description for brevity) to a type graph
3914  * formed by (potential) canonical struct/union ("canonical graph" for brevity
3915  * here, though keep in mind that not all types in canonical graph are
3916  * necessarily canonical representatives themselves, some of them might be
3917  * duplicates or its uniqueness might not have been established yet).
3918  * Returns:
3919  *  - >0, if type graphs are equivalent;
3920  *  -  0, if not equivalent;
3921  *  - <0, on error.
3922  *
3923  * Algorithm performs side-by-side DFS traversal of both type graphs and checks
3924  * equivalence of BTF types at each step. If at any point BTF types in candidate
3925  * and canonical graphs are not compatible structurally, whole graphs are
3926  * incompatible. If types are structurally equivalent (i.e., all information
3927  * except referenced type IDs is exactly the same), a mapping from `canon_id` to
3928  * a `cand_id` is recored in hypothetical mapping (`btf_dedup->hypot_map`).
3929  * If a type references other types, then those referenced types are checked
3930  * for equivalence recursively.
3931  *
3932  * During DFS traversal, if we find that for current `canon_id` type we
3933  * already have some mapping in hypothetical map, we check for two possible
3934  * situations:
3935  *   - `canon_id` is mapped to exactly the same type as `cand_id`. This will
3936  *     happen when type graphs have cycles. In this case we assume those two
3937  *     types are equivalent.
3938  *   - `canon_id` is mapped to different type. This is contradiction in our
3939  *     hypothetical mapping, because same graph in canonical graph corresponds
3940  *     to two different types in candidate graph, which for equivalent type
3941  *     graphs shouldn't happen. This condition terminates equivalence check
3942  *     with negative result.
3943  *
3944  * If type graphs traversal exhausts types to check and find no contradiction,
3945  * then type graphs are equivalent.
3946  *
3947  * When checking types for equivalence, there is one special case: FWD types.
3948  * If FWD type resolution is allowed and one of the types (either from canonical
3949  * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
3950  * flag) and their names match, hypothetical mapping is updated to point from
3951  * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
3952  * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
3953  *
3954  * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
3955  * if there are two exactly named (or anonymous) structs/unions that are
3956  * compatible structurally, one of which has FWD field, while other is concrete
3957  * STRUCT/UNION, but according to C sources they are different structs/unions
3958  * that are referencing different types with the same name. This is extremely
3959  * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
3960  * this logic is causing problems.
3961  *
3962  * Doing FWD resolution means that both candidate and/or canonical graphs can
3963  * consists of portions of the graph that come from multiple compilation units.
3964  * This is due to the fact that types within single compilation unit are always
3965  * deduplicated and FWDs are already resolved, if referenced struct/union
3966  * definiton is available. So, if we had unresolved FWD and found corresponding
3967  * STRUCT/UNION, they will be from different compilation units. This
3968  * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
3969  * type graph will likely have at least two different BTF types that describe
3970  * same type (e.g., most probably there will be two different BTF types for the
3971  * same 'int' primitive type) and could even have "overlapping" parts of type
3972  * graph that describe same subset of types.
3973  *
3974  * This in turn means that our assumption that each type in canonical graph
3975  * must correspond to exactly one type in candidate graph might not hold
3976  * anymore and will make it harder to detect contradictions using hypothetical
3977  * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
3978  * resolution only in canonical graph. FWDs in candidate graphs are never
3979  * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
3980  * that can occur:
3981  *   - Both types in canonical and candidate graphs are FWDs. If they are
3982  *     structurally equivalent, then they can either be both resolved to the
3983  *     same STRUCT/UNION or not resolved at all. In both cases they are
3984  *     equivalent and there is no need to resolve FWD on candidate side.
3985  *   - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
3986  *     so nothing to resolve as well, algorithm will check equivalence anyway.
3987  *   - Type in canonical graph is FWD, while type in candidate is concrete
3988  *     STRUCT/UNION. In this case candidate graph comes from single compilation
3989  *     unit, so there is exactly one BTF type for each unique C type. After
3990  *     resolving FWD into STRUCT/UNION, there might be more than one BTF type
3991  *     in canonical graph mapping to single BTF type in candidate graph, but
3992  *     because hypothetical mapping maps from canonical to candidate types, it's
3993  *     alright, and we still maintain the property of having single `canon_id`
3994  *     mapping to single `cand_id` (there could be two different `canon_id`
3995  *     mapped to the same `cand_id`, but it's not contradictory).
3996  *   - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
3997  *     graph is FWD. In this case we are just going to check compatibility of
3998  *     STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
3999  *     assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
4000  *     a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
4001  *     turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
4002  *     canonical graph.
4003  */
4004 static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
4005 			      __u32 canon_id)
4006 {
4007 	struct btf_type *cand_type;
4008 	struct btf_type *canon_type;
4009 	__u32 hypot_type_id;
4010 	__u16 cand_kind;
4011 	__u16 canon_kind;
4012 	int i, eq;
4013 
4014 	/* if both resolve to the same canonical, they must be equivalent */
4015 	if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
4016 		return 1;
4017 
4018 	canon_id = resolve_fwd_id(d, canon_id);
4019 
4020 	hypot_type_id = d->hypot_map[canon_id];
4021 	if (hypot_type_id <= BTF_MAX_NR_TYPES) {
4022 		if (hypot_type_id == cand_id)
4023 			return 1;
4024 		/* In some cases compiler will generate different DWARF types
4025 		 * for *identical* array type definitions and use them for
4026 		 * different fields within the *same* struct. This breaks type
4027 		 * equivalence check, which makes an assumption that candidate
4028 		 * types sub-graph has a consistent and deduped-by-compiler
4029 		 * types within a single CU. So work around that by explicitly
4030 		 * allowing identical array types here.
4031 		 */
4032 		if (btf_dedup_identical_arrays(d, hypot_type_id, cand_id))
4033 			return 1;
4034 		/* It turns out that similar situation can happen with
4035 		 * struct/union sometimes, sigh... Handle the case where
4036 		 * structs/unions are exactly the same, down to the referenced
4037 		 * type IDs. Anything more complicated (e.g., if referenced
4038 		 * types are different, but equivalent) is *way more*
4039 		 * complicated and requires a many-to-many equivalence mapping.
4040 		 */
4041 		if (btf_dedup_identical_structs(d, hypot_type_id, cand_id))
4042 			return 1;
4043 		return 0;
4044 	}
4045 
4046 	if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
4047 		return -ENOMEM;
4048 
4049 	cand_type = btf_type_by_id(d->btf, cand_id);
4050 	canon_type = btf_type_by_id(d->btf, canon_id);
4051 	cand_kind = btf_kind(cand_type);
4052 	canon_kind = btf_kind(canon_type);
4053 
4054 	if (cand_type->name_off != canon_type->name_off)
4055 		return 0;
4056 
4057 	/* FWD <--> STRUCT/UNION equivalence check, if enabled */
4058 	if ((cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
4059 	    && cand_kind != canon_kind) {
4060 		__u16 real_kind;
4061 		__u16 fwd_kind;
4062 
4063 		if (cand_kind == BTF_KIND_FWD) {
4064 			real_kind = canon_kind;
4065 			fwd_kind = btf_fwd_kind(cand_type);
4066 		} else {
4067 			real_kind = cand_kind;
4068 			fwd_kind = btf_fwd_kind(canon_type);
4069 			/* we'd need to resolve base FWD to STRUCT/UNION */
4070 			if (fwd_kind == real_kind && canon_id < d->btf->start_id)
4071 				d->hypot_adjust_canon = true;
4072 		}
4073 		return fwd_kind == real_kind;
4074 	}
4075 
4076 	if (cand_kind != canon_kind)
4077 		return 0;
4078 
4079 	switch (cand_kind) {
4080 	case BTF_KIND_INT:
4081 		return btf_equal_int_tag(cand_type, canon_type);
4082 
4083 	case BTF_KIND_ENUM:
4084 		return btf_compat_enum(cand_type, canon_type);
4085 
4086 	case BTF_KIND_FWD:
4087 	case BTF_KIND_FLOAT:
4088 		return btf_equal_common(cand_type, canon_type);
4089 
4090 	case BTF_KIND_CONST:
4091 	case BTF_KIND_VOLATILE:
4092 	case BTF_KIND_RESTRICT:
4093 	case BTF_KIND_PTR:
4094 	case BTF_KIND_TYPEDEF:
4095 	case BTF_KIND_FUNC:
4096 	case BTF_KIND_TYPE_TAG:
4097 		if (cand_type->info != canon_type->info)
4098 			return 0;
4099 		return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4100 
4101 	case BTF_KIND_ARRAY: {
4102 		const struct btf_array *cand_arr, *canon_arr;
4103 
4104 		if (!btf_compat_array(cand_type, canon_type))
4105 			return 0;
4106 		cand_arr = btf_array(cand_type);
4107 		canon_arr = btf_array(canon_type);
4108 		eq = btf_dedup_is_equiv(d, cand_arr->index_type, canon_arr->index_type);
4109 		if (eq <= 0)
4110 			return eq;
4111 		return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
4112 	}
4113 
4114 	case BTF_KIND_STRUCT:
4115 	case BTF_KIND_UNION: {
4116 		const struct btf_member *cand_m, *canon_m;
4117 		__u16 vlen;
4118 
4119 		if (!btf_shallow_equal_struct(cand_type, canon_type))
4120 			return 0;
4121 		vlen = btf_vlen(cand_type);
4122 		cand_m = btf_members(cand_type);
4123 		canon_m = btf_members(canon_type);
4124 		for (i = 0; i < vlen; i++) {
4125 			eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
4126 			if (eq <= 0)
4127 				return eq;
4128 			cand_m++;
4129 			canon_m++;
4130 		}
4131 
4132 		return 1;
4133 	}
4134 
4135 	case BTF_KIND_FUNC_PROTO: {
4136 		const struct btf_param *cand_p, *canon_p;
4137 		__u16 vlen;
4138 
4139 		if (!btf_compat_fnproto(cand_type, canon_type))
4140 			return 0;
4141 		eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4142 		if (eq <= 0)
4143 			return eq;
4144 		vlen = btf_vlen(cand_type);
4145 		cand_p = btf_params(cand_type);
4146 		canon_p = btf_params(canon_type);
4147 		for (i = 0; i < vlen; i++) {
4148 			eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
4149 			if (eq <= 0)
4150 				return eq;
4151 			cand_p++;
4152 			canon_p++;
4153 		}
4154 		return 1;
4155 	}
4156 
4157 	default:
4158 		return -EINVAL;
4159 	}
4160 	return 0;
4161 }
4162 
4163 /*
4164  * Use hypothetical mapping, produced by successful type graph equivalence
4165  * check, to augment existing struct/union canonical mapping, where possible.
4166  *
4167  * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
4168  * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
4169  * it doesn't matter if FWD type was part of canonical graph or candidate one,
4170  * we are recording the mapping anyway. As opposed to carefulness required
4171  * for struct/union correspondence mapping (described below), for FWD resolution
4172  * it's not important, as by the time that FWD type (reference type) will be
4173  * deduplicated all structs/unions will be deduped already anyway.
4174  *
4175  * Recording STRUCT/UNION mapping is purely a performance optimization and is
4176  * not required for correctness. It needs to be done carefully to ensure that
4177  * struct/union from candidate's type graph is not mapped into corresponding
4178  * struct/union from canonical type graph that itself hasn't been resolved into
4179  * canonical representative. The only guarantee we have is that canonical
4180  * struct/union was determined as canonical and that won't change. But any
4181  * types referenced through that struct/union fields could have been not yet
4182  * resolved, so in case like that it's too early to establish any kind of
4183  * correspondence between structs/unions.
4184  *
4185  * No canonical correspondence is derived for primitive types (they are already
4186  * deduplicated completely already anyway) or reference types (they rely on
4187  * stability of struct/union canonical relationship for equivalence checks).
4188  */
4189 static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
4190 {
4191 	__u32 canon_type_id, targ_type_id;
4192 	__u16 t_kind, c_kind;
4193 	__u32 t_id, c_id;
4194 	int i;
4195 
4196 	for (i = 0; i < d->hypot_cnt; i++) {
4197 		canon_type_id = d->hypot_list[i];
4198 		targ_type_id = d->hypot_map[canon_type_id];
4199 		t_id = resolve_type_id(d, targ_type_id);
4200 		c_id = resolve_type_id(d, canon_type_id);
4201 		t_kind = btf_kind(btf__type_by_id(d->btf, t_id));
4202 		c_kind = btf_kind(btf__type_by_id(d->btf, c_id));
4203 		/*
4204 		 * Resolve FWD into STRUCT/UNION.
4205 		 * It's ok to resolve FWD into STRUCT/UNION that's not yet
4206 		 * mapped to canonical representative (as opposed to
4207 		 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
4208 		 * eventually that struct is going to be mapped and all resolved
4209 		 * FWDs will automatically resolve to correct canonical
4210 		 * representative. This will happen before ref type deduping,
4211 		 * which critically depends on stability of these mapping. This
4212 		 * stability is not a requirement for STRUCT/UNION equivalence
4213 		 * checks, though.
4214 		 */
4215 
4216 		/* if it's the split BTF case, we still need to point base FWD
4217 		 * to STRUCT/UNION in a split BTF, because FWDs from split BTF
4218 		 * will be resolved against base FWD. If we don't point base
4219 		 * canonical FWD to the resolved STRUCT/UNION, then all the
4220 		 * FWDs in split BTF won't be correctly resolved to a proper
4221 		 * STRUCT/UNION.
4222 		 */
4223 		if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
4224 			d->map[c_id] = t_id;
4225 
4226 		/* if graph equivalence determined that we'd need to adjust
4227 		 * base canonical types, then we need to only point base FWDs
4228 		 * to STRUCTs/UNIONs and do no more modifications. For all
4229 		 * other purposes the type graphs were not equivalent.
4230 		 */
4231 		if (d->hypot_adjust_canon)
4232 			continue;
4233 
4234 		if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
4235 			d->map[t_id] = c_id;
4236 
4237 		if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
4238 		    c_kind != BTF_KIND_FWD &&
4239 		    is_type_mapped(d, c_id) &&
4240 		    !is_type_mapped(d, t_id)) {
4241 			/*
4242 			 * as a perf optimization, we can map struct/union
4243 			 * that's part of type graph we just verified for
4244 			 * equivalence. We can do that for struct/union that has
4245 			 * canonical representative only, though.
4246 			 */
4247 			d->map[t_id] = c_id;
4248 		}
4249 	}
4250 }
4251 
4252 /*
4253  * Deduplicate struct/union types.
4254  *
4255  * For each struct/union type its type signature hash is calculated, taking
4256  * into account type's name, size, number, order and names of fields, but
4257  * ignoring type ID's referenced from fields, because they might not be deduped
4258  * completely until after reference types deduplication phase. This type hash
4259  * is used to iterate over all potential canonical types, sharing same hash.
4260  * For each canonical candidate we check whether type graphs that they form
4261  * (through referenced types in fields and so on) are equivalent using algorithm
4262  * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
4263  * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
4264  * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
4265  * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
4266  * potentially map other structs/unions to their canonical representatives,
4267  * if such relationship hasn't yet been established. This speeds up algorithm
4268  * by eliminating some of the duplicate work.
4269  *
4270  * If no matching canonical representative was found, struct/union is marked
4271  * as canonical for itself and is added into btf_dedup->dedup_table hash map
4272  * for further look ups.
4273  */
4274 static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
4275 {
4276 	struct btf_type *cand_type, *t;
4277 	struct hashmap_entry *hash_entry;
4278 	/* if we don't find equivalent type, then we are canonical */
4279 	__u32 new_id = type_id;
4280 	__u16 kind;
4281 	long h;
4282 
4283 	/* already deduped or is in process of deduping (loop detected) */
4284 	if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4285 		return 0;
4286 
4287 	t = btf_type_by_id(d->btf, type_id);
4288 	kind = btf_kind(t);
4289 
4290 	if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4291 		return 0;
4292 
4293 	h = btf_hash_struct(t);
4294 	for_each_dedup_cand(d, hash_entry, h) {
4295 		__u32 cand_id = (__u32)(long)hash_entry->value;
4296 		int eq;
4297 
4298 		/*
4299 		 * Even though btf_dedup_is_equiv() checks for
4300 		 * btf_shallow_equal_struct() internally when checking two
4301 		 * structs (unions) for equivalence, we need to guard here
4302 		 * from picking matching FWD type as a dedup candidate.
4303 		 * This can happen due to hash collision. In such case just
4304 		 * relying on btf_dedup_is_equiv() would lead to potentially
4305 		 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
4306 		 * FWD and compatible STRUCT/UNION are considered equivalent.
4307 		 */
4308 		cand_type = btf_type_by_id(d->btf, cand_id);
4309 		if (!btf_shallow_equal_struct(t, cand_type))
4310 			continue;
4311 
4312 		btf_dedup_clear_hypot_map(d);
4313 		eq = btf_dedup_is_equiv(d, type_id, cand_id);
4314 		if (eq < 0)
4315 			return eq;
4316 		if (!eq)
4317 			continue;
4318 		btf_dedup_merge_hypot_map(d);
4319 		if (d->hypot_adjust_canon) /* not really equivalent */
4320 			continue;
4321 		new_id = cand_id;
4322 		break;
4323 	}
4324 
4325 	d->map[type_id] = new_id;
4326 	if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4327 		return -ENOMEM;
4328 
4329 	return 0;
4330 }
4331 
4332 static int btf_dedup_struct_types(struct btf_dedup *d)
4333 {
4334 	int i, err;
4335 
4336 	for (i = 0; i < d->btf->nr_types; i++) {
4337 		err = btf_dedup_struct_type(d, d->btf->start_id + i);
4338 		if (err)
4339 			return err;
4340 	}
4341 	return 0;
4342 }
4343 
4344 /*
4345  * Deduplicate reference type.
4346  *
4347  * Once all primitive and struct/union types got deduplicated, we can easily
4348  * deduplicate all other (reference) BTF types. This is done in two steps:
4349  *
4350  * 1. Resolve all referenced type IDs into their canonical type IDs. This
4351  * resolution can be done either immediately for primitive or struct/union types
4352  * (because they were deduped in previous two phases) or recursively for
4353  * reference types. Recursion will always terminate at either primitive or
4354  * struct/union type, at which point we can "unwind" chain of reference types
4355  * one by one. There is no danger of encountering cycles because in C type
4356  * system the only way to form type cycle is through struct/union, so any chain
4357  * of reference types, even those taking part in a type cycle, will inevitably
4358  * reach struct/union at some point.
4359  *
4360  * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
4361  * becomes "stable", in the sense that no further deduplication will cause
4362  * any changes to it. With that, it's now possible to calculate type's signature
4363  * hash (this time taking into account referenced type IDs) and loop over all
4364  * potential canonical representatives. If no match was found, current type
4365  * will become canonical representative of itself and will be added into
4366  * btf_dedup->dedup_table as another possible canonical representative.
4367  */
4368 static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
4369 {
4370 	struct hashmap_entry *hash_entry;
4371 	__u32 new_id = type_id, cand_id;
4372 	struct btf_type *t, *cand;
4373 	/* if we don't find equivalent type, then we are representative type */
4374 	int ref_type_id;
4375 	long h;
4376 
4377 	if (d->map[type_id] == BTF_IN_PROGRESS_ID)
4378 		return -ELOOP;
4379 	if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4380 		return resolve_type_id(d, type_id);
4381 
4382 	t = btf_type_by_id(d->btf, type_id);
4383 	d->map[type_id] = BTF_IN_PROGRESS_ID;
4384 
4385 	switch (btf_kind(t)) {
4386 	case BTF_KIND_CONST:
4387 	case BTF_KIND_VOLATILE:
4388 	case BTF_KIND_RESTRICT:
4389 	case BTF_KIND_PTR:
4390 	case BTF_KIND_TYPEDEF:
4391 	case BTF_KIND_FUNC:
4392 	case BTF_KIND_TYPE_TAG:
4393 		ref_type_id = btf_dedup_ref_type(d, t->type);
4394 		if (ref_type_id < 0)
4395 			return ref_type_id;
4396 		t->type = ref_type_id;
4397 
4398 		h = btf_hash_common(t);
4399 		for_each_dedup_cand(d, hash_entry, h) {
4400 			cand_id = (__u32)(long)hash_entry->value;
4401 			cand = btf_type_by_id(d->btf, cand_id);
4402 			if (btf_equal_common(t, cand)) {
4403 				new_id = cand_id;
4404 				break;
4405 			}
4406 		}
4407 		break;
4408 
4409 	case BTF_KIND_DECL_TAG:
4410 		ref_type_id = btf_dedup_ref_type(d, t->type);
4411 		if (ref_type_id < 0)
4412 			return ref_type_id;
4413 		t->type = ref_type_id;
4414 
4415 		h = btf_hash_int_decl_tag(t);
4416 		for_each_dedup_cand(d, hash_entry, h) {
4417 			cand_id = (__u32)(long)hash_entry->value;
4418 			cand = btf_type_by_id(d->btf, cand_id);
4419 			if (btf_equal_int_tag(t, cand)) {
4420 				new_id = cand_id;
4421 				break;
4422 			}
4423 		}
4424 		break;
4425 
4426 	case BTF_KIND_ARRAY: {
4427 		struct btf_array *info = btf_array(t);
4428 
4429 		ref_type_id = btf_dedup_ref_type(d, info->type);
4430 		if (ref_type_id < 0)
4431 			return ref_type_id;
4432 		info->type = ref_type_id;
4433 
4434 		ref_type_id = btf_dedup_ref_type(d, info->index_type);
4435 		if (ref_type_id < 0)
4436 			return ref_type_id;
4437 		info->index_type = ref_type_id;
4438 
4439 		h = btf_hash_array(t);
4440 		for_each_dedup_cand(d, hash_entry, h) {
4441 			cand_id = (__u32)(long)hash_entry->value;
4442 			cand = btf_type_by_id(d->btf, cand_id);
4443 			if (btf_equal_array(t, cand)) {
4444 				new_id = cand_id;
4445 				break;
4446 			}
4447 		}
4448 		break;
4449 	}
4450 
4451 	case BTF_KIND_FUNC_PROTO: {
4452 		struct btf_param *param;
4453 		__u16 vlen;
4454 		int i;
4455 
4456 		ref_type_id = btf_dedup_ref_type(d, t->type);
4457 		if (ref_type_id < 0)
4458 			return ref_type_id;
4459 		t->type = ref_type_id;
4460 
4461 		vlen = btf_vlen(t);
4462 		param = btf_params(t);
4463 		for (i = 0; i < vlen; i++) {
4464 			ref_type_id = btf_dedup_ref_type(d, param->type);
4465 			if (ref_type_id < 0)
4466 				return ref_type_id;
4467 			param->type = ref_type_id;
4468 			param++;
4469 		}
4470 
4471 		h = btf_hash_fnproto(t);
4472 		for_each_dedup_cand(d, hash_entry, h) {
4473 			cand_id = (__u32)(long)hash_entry->value;
4474 			cand = btf_type_by_id(d->btf, cand_id);
4475 			if (btf_equal_fnproto(t, cand)) {
4476 				new_id = cand_id;
4477 				break;
4478 			}
4479 		}
4480 		break;
4481 	}
4482 
4483 	default:
4484 		return -EINVAL;
4485 	}
4486 
4487 	d->map[type_id] = new_id;
4488 	if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4489 		return -ENOMEM;
4490 
4491 	return new_id;
4492 }
4493 
4494 static int btf_dedup_ref_types(struct btf_dedup *d)
4495 {
4496 	int i, err;
4497 
4498 	for (i = 0; i < d->btf->nr_types; i++) {
4499 		err = btf_dedup_ref_type(d, d->btf->start_id + i);
4500 		if (err < 0)
4501 			return err;
4502 	}
4503 	/* we won't need d->dedup_table anymore */
4504 	hashmap__free(d->dedup_table);
4505 	d->dedup_table = NULL;
4506 	return 0;
4507 }
4508 
4509 /*
4510  * Compact types.
4511  *
4512  * After we established for each type its corresponding canonical representative
4513  * type, we now can eliminate types that are not canonical and leave only
4514  * canonical ones layed out sequentially in memory by copying them over
4515  * duplicates. During compaction btf_dedup->hypot_map array is reused to store
4516  * a map from original type ID to a new compacted type ID, which will be used
4517  * during next phase to "fix up" type IDs, referenced from struct/union and
4518  * reference types.
4519  */
4520 static int btf_dedup_compact_types(struct btf_dedup *d)
4521 {
4522 	__u32 *new_offs;
4523 	__u32 next_type_id = d->btf->start_id;
4524 	const struct btf_type *t;
4525 	void *p;
4526 	int i, id, len;
4527 
4528 	/* we are going to reuse hypot_map to store compaction remapping */
4529 	d->hypot_map[0] = 0;
4530 	/* base BTF types are not renumbered */
4531 	for (id = 1; id < d->btf->start_id; id++)
4532 		d->hypot_map[id] = id;
4533 	for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++)
4534 		d->hypot_map[id] = BTF_UNPROCESSED_ID;
4535 
4536 	p = d->btf->types_data;
4537 
4538 	for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++) {
4539 		if (d->map[id] != id)
4540 			continue;
4541 
4542 		t = btf__type_by_id(d->btf, id);
4543 		len = btf_type_size(t);
4544 		if (len < 0)
4545 			return len;
4546 
4547 		memmove(p, t, len);
4548 		d->hypot_map[id] = next_type_id;
4549 		d->btf->type_offs[next_type_id - d->btf->start_id] = p - d->btf->types_data;
4550 		p += len;
4551 		next_type_id++;
4552 	}
4553 
4554 	/* shrink struct btf's internal types index and update btf_header */
4555 	d->btf->nr_types = next_type_id - d->btf->start_id;
4556 	d->btf->type_offs_cap = d->btf->nr_types;
4557 	d->btf->hdr->type_len = p - d->btf->types_data;
4558 	new_offs = libbpf_reallocarray(d->btf->type_offs, d->btf->type_offs_cap,
4559 				       sizeof(*new_offs));
4560 	if (d->btf->type_offs_cap && !new_offs)
4561 		return -ENOMEM;
4562 	d->btf->type_offs = new_offs;
4563 	d->btf->hdr->str_off = d->btf->hdr->type_len;
4564 	d->btf->raw_size = d->btf->hdr->hdr_len + d->btf->hdr->type_len + d->btf->hdr->str_len;
4565 	return 0;
4566 }
4567 
4568 /*
4569  * Figure out final (deduplicated and compacted) type ID for provided original
4570  * `type_id` by first resolving it into corresponding canonical type ID and
4571  * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
4572  * which is populated during compaction phase.
4573  */
4574 static int btf_dedup_remap_type_id(__u32 *type_id, void *ctx)
4575 {
4576 	struct btf_dedup *d = ctx;
4577 	__u32 resolved_type_id, new_type_id;
4578 
4579 	resolved_type_id = resolve_type_id(d, *type_id);
4580 	new_type_id = d->hypot_map[resolved_type_id];
4581 	if (new_type_id > BTF_MAX_NR_TYPES)
4582 		return -EINVAL;
4583 
4584 	*type_id = new_type_id;
4585 	return 0;
4586 }
4587 
4588 /*
4589  * Remap referenced type IDs into deduped type IDs.
4590  *
4591  * After BTF types are deduplicated and compacted, their final type IDs may
4592  * differ from original ones. The map from original to a corresponding
4593  * deduped type ID is stored in btf_dedup->hypot_map and is populated during
4594  * compaction phase. During remapping phase we are rewriting all type IDs
4595  * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
4596  * their final deduped type IDs.
4597  */
4598 static int btf_dedup_remap_types(struct btf_dedup *d)
4599 {
4600 	int i, r;
4601 
4602 	for (i = 0; i < d->btf->nr_types; i++) {
4603 		struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
4604 
4605 		r = btf_type_visit_type_ids(t, btf_dedup_remap_type_id, d);
4606 		if (r)
4607 			return r;
4608 	}
4609 
4610 	if (!d->btf_ext)
4611 		return 0;
4612 
4613 	r = btf_ext_visit_type_ids(d->btf_ext, btf_dedup_remap_type_id, d);
4614 	if (r)
4615 		return r;
4616 
4617 	return 0;
4618 }
4619 
4620 /*
4621  * Probe few well-known locations for vmlinux kernel image and try to load BTF
4622  * data out of it to use for target BTF.
4623  */
4624 struct btf *btf__load_vmlinux_btf(void)
4625 {
4626 	struct {
4627 		const char *path_fmt;
4628 		bool raw_btf;
4629 	} locations[] = {
4630 		/* try canonical vmlinux BTF through sysfs first */
4631 		{ "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
4632 		/* fall back to trying to find vmlinux ELF on disk otherwise */
4633 		{ "/boot/vmlinux-%1$s" },
4634 		{ "/lib/modules/%1$s/vmlinux-%1$s" },
4635 		{ "/lib/modules/%1$s/build/vmlinux" },
4636 		{ "/usr/lib/modules/%1$s/kernel/vmlinux" },
4637 		{ "/usr/lib/debug/boot/vmlinux-%1$s" },
4638 		{ "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
4639 		{ "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
4640 	};
4641 	char path[PATH_MAX + 1];
4642 	struct utsname buf;
4643 	struct btf *btf;
4644 	int i, err;
4645 
4646 	uname(&buf);
4647 
4648 	for (i = 0; i < ARRAY_SIZE(locations); i++) {
4649 		snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
4650 
4651 		if (access(path, R_OK))
4652 			continue;
4653 
4654 		if (locations[i].raw_btf)
4655 			btf = btf__parse_raw(path);
4656 		else
4657 			btf = btf__parse_elf(path, NULL);
4658 		err = libbpf_get_error(btf);
4659 		pr_debug("loading kernel BTF '%s': %d\n", path, err);
4660 		if (err)
4661 			continue;
4662 
4663 		return btf;
4664 	}
4665 
4666 	pr_warn("failed to find valid kernel BTF\n");
4667 	return libbpf_err_ptr(-ESRCH);
4668 }
4669 
4670 struct btf *libbpf_find_kernel_btf(void) __attribute__((alias("btf__load_vmlinux_btf")));
4671 
4672 struct btf *btf__load_module_btf(const char *module_name, struct btf *vmlinux_btf)
4673 {
4674 	char path[80];
4675 
4676 	snprintf(path, sizeof(path), "/sys/kernel/btf/%s", module_name);
4677 	return btf__parse_split(path, vmlinux_btf);
4678 }
4679 
4680 int btf_type_visit_type_ids(struct btf_type *t, type_id_visit_fn visit, void *ctx)
4681 {
4682 	int i, n, err;
4683 
4684 	switch (btf_kind(t)) {
4685 	case BTF_KIND_INT:
4686 	case BTF_KIND_FLOAT:
4687 	case BTF_KIND_ENUM:
4688 		return 0;
4689 
4690 	case BTF_KIND_FWD:
4691 	case BTF_KIND_CONST:
4692 	case BTF_KIND_VOLATILE:
4693 	case BTF_KIND_RESTRICT:
4694 	case BTF_KIND_PTR:
4695 	case BTF_KIND_TYPEDEF:
4696 	case BTF_KIND_FUNC:
4697 	case BTF_KIND_VAR:
4698 	case BTF_KIND_DECL_TAG:
4699 	case BTF_KIND_TYPE_TAG:
4700 		return visit(&t->type, ctx);
4701 
4702 	case BTF_KIND_ARRAY: {
4703 		struct btf_array *a = btf_array(t);
4704 
4705 		err = visit(&a->type, ctx);
4706 		err = err ?: visit(&a->index_type, ctx);
4707 		return err;
4708 	}
4709 
4710 	case BTF_KIND_STRUCT:
4711 	case BTF_KIND_UNION: {
4712 		struct btf_member *m = btf_members(t);
4713 
4714 		for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4715 			err = visit(&m->type, ctx);
4716 			if (err)
4717 				return err;
4718 		}
4719 		return 0;
4720 	}
4721 
4722 	case BTF_KIND_FUNC_PROTO: {
4723 		struct btf_param *m = btf_params(t);
4724 
4725 		err = visit(&t->type, ctx);
4726 		if (err)
4727 			return err;
4728 		for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4729 			err = visit(&m->type, ctx);
4730 			if (err)
4731 				return err;
4732 		}
4733 		return 0;
4734 	}
4735 
4736 	case BTF_KIND_DATASEC: {
4737 		struct btf_var_secinfo *m = btf_var_secinfos(t);
4738 
4739 		for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4740 			err = visit(&m->type, ctx);
4741 			if (err)
4742 				return err;
4743 		}
4744 		return 0;
4745 	}
4746 
4747 	default:
4748 		return -EINVAL;
4749 	}
4750 }
4751 
4752 int btf_type_visit_str_offs(struct btf_type *t, str_off_visit_fn visit, void *ctx)
4753 {
4754 	int i, n, err;
4755 
4756 	err = visit(&t->name_off, ctx);
4757 	if (err)
4758 		return err;
4759 
4760 	switch (btf_kind(t)) {
4761 	case BTF_KIND_STRUCT:
4762 	case BTF_KIND_UNION: {
4763 		struct btf_member *m = btf_members(t);
4764 
4765 		for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4766 			err = visit(&m->name_off, ctx);
4767 			if (err)
4768 				return err;
4769 		}
4770 		break;
4771 	}
4772 	case BTF_KIND_ENUM: {
4773 		struct btf_enum *m = btf_enum(t);
4774 
4775 		for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4776 			err = visit(&m->name_off, ctx);
4777 			if (err)
4778 				return err;
4779 		}
4780 		break;
4781 	}
4782 	case BTF_KIND_FUNC_PROTO: {
4783 		struct btf_param *m = btf_params(t);
4784 
4785 		for (i = 0, n = btf_vlen(t); i < n; i++, m++) {
4786 			err = visit(&m->name_off, ctx);
4787 			if (err)
4788 				return err;
4789 		}
4790 		break;
4791 	}
4792 	default:
4793 		break;
4794 	}
4795 
4796 	return 0;
4797 }
4798 
4799 int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx)
4800 {
4801 	const struct btf_ext_info *seg;
4802 	struct btf_ext_info_sec *sec;
4803 	int i, err;
4804 
4805 	seg = &btf_ext->func_info;
4806 	for_each_btf_ext_sec(seg, sec) {
4807 		struct bpf_func_info_min *rec;
4808 
4809 		for_each_btf_ext_rec(seg, sec, i, rec) {
4810 			err = visit(&rec->type_id, ctx);
4811 			if (err < 0)
4812 				return err;
4813 		}
4814 	}
4815 
4816 	seg = &btf_ext->core_relo_info;
4817 	for_each_btf_ext_sec(seg, sec) {
4818 		struct bpf_core_relo *rec;
4819 
4820 		for_each_btf_ext_rec(seg, sec, i, rec) {
4821 			err = visit(&rec->type_id, ctx);
4822 			if (err < 0)
4823 				return err;
4824 		}
4825 	}
4826 
4827 	return 0;
4828 }
4829 
4830 int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx)
4831 {
4832 	const struct btf_ext_info *seg;
4833 	struct btf_ext_info_sec *sec;
4834 	int i, err;
4835 
4836 	seg = &btf_ext->func_info;
4837 	for_each_btf_ext_sec(seg, sec) {
4838 		err = visit(&sec->sec_name_off, ctx);
4839 		if (err)
4840 			return err;
4841 	}
4842 
4843 	seg = &btf_ext->line_info;
4844 	for_each_btf_ext_sec(seg, sec) {
4845 		struct bpf_line_info_min *rec;
4846 
4847 		err = visit(&sec->sec_name_off, ctx);
4848 		if (err)
4849 			return err;
4850 
4851 		for_each_btf_ext_rec(seg, sec, i, rec) {
4852 			err = visit(&rec->file_name_off, ctx);
4853 			if (err)
4854 				return err;
4855 			err = visit(&rec->line_off, ctx);
4856 			if (err)
4857 				return err;
4858 		}
4859 	}
4860 
4861 	seg = &btf_ext->core_relo_info;
4862 	for_each_btf_ext_sec(seg, sec) {
4863 		struct bpf_core_relo *rec;
4864 
4865 		err = visit(&sec->sec_name_off, ctx);
4866 		if (err)
4867 			return err;
4868 
4869 		for_each_btf_ext_rec(seg, sec, i, rec) {
4870 			err = visit(&rec->access_str_off, ctx);
4871 			if (err)
4872 				return err;
4873 		}
4874 	}
4875 
4876 	return 0;
4877 }
4878