xref: /linux/tools/lib/bpf/btf.c (revision 38c6104e0bc7c8af20ab4897cb0504e3339e4fe4)
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 #include "str_error.h"
26 
27 #define BTF_MAX_NR_TYPES 0x7fffffffU
28 #define BTF_MAX_STR_OFFSET 0x7fffffffU
29 
30 static struct btf_type btf_void;
31 
32 struct btf {
33 	/* raw BTF data in native endianness */
34 	void *raw_data;
35 	/* raw BTF data in non-native endianness */
36 	void *raw_data_swapped;
37 	__u32 raw_size;
38 	/* whether target endianness differs from the native one */
39 	bool swapped_endian;
40 
41 	/*
42 	 * When BTF is loaded from an ELF or raw memory it is stored
43 	 * in a contiguous memory block. The hdr, type_data, and, strs_data
44 	 * point inside that memory region to their respective parts of BTF
45 	 * representation:
46 	 *
47 	 * +--------------------------------+
48 	 * |  Header  |  Types  |  Strings  |
49 	 * +--------------------------------+
50 	 * ^          ^         ^
51 	 * |          |         |
52 	 * hdr        |         |
53 	 * types_data-+         |
54 	 * strs_data------------+
55 	 *
56 	 * If BTF data is later modified, e.g., due to types added or
57 	 * removed, BTF deduplication performed, etc, this contiguous
58 	 * representation is broken up into three independently allocated
59 	 * memory regions to be able to modify them independently.
60 	 * raw_data is nulled out at that point, but can be later allocated
61 	 * and cached again if user calls btf__raw_data(), at which point
62 	 * raw_data will contain a contiguous copy of header, types, and
63 	 * strings:
64 	 *
65 	 * +----------+  +---------+  +-----------+
66 	 * |  Header  |  |  Types  |  |  Strings  |
67 	 * +----------+  +---------+  +-----------+
68 	 * ^             ^            ^
69 	 * |             |            |
70 	 * hdr           |            |
71 	 * types_data----+            |
72 	 * strset__data(strs_set)-----+
73 	 *
74 	 *               +----------+---------+-----------+
75 	 *               |  Header  |  Types  |  Strings  |
76 	 * raw_data----->+----------+---------+-----------+
77 	 */
78 	struct btf_header *hdr;
79 
80 	void *types_data;
81 	size_t types_data_cap; /* used size stored in hdr->type_len */
82 
83 	/* type ID to `struct btf_type *` lookup index
84 	 * type_offs[0] corresponds to the first non-VOID type:
85 	 *   - for base BTF it's type [1];
86 	 *   - for split BTF it's the first non-base BTF type.
87 	 */
88 	__u32 *type_offs;
89 	size_t type_offs_cap;
90 	/* number of types in this BTF instance:
91 	 *   - doesn't include special [0] void type;
92 	 *   - for split BTF counts number of types added on top of base BTF.
93 	 */
94 	__u32 nr_types;
95 	/* if not NULL, points to the base BTF on top of which the current
96 	 * split BTF is based
97 	 */
98 	struct btf *base_btf;
99 	/* BTF type ID of the first type in this BTF instance:
100 	 *   - for base BTF it's equal to 1;
101 	 *   - for split BTF it's equal to biggest type ID of base BTF plus 1.
102 	 */
103 	int start_id;
104 	/* logical string offset of this BTF instance:
105 	 *   - for base BTF it's equal to 0;
106 	 *   - for split BTF it's equal to total size of base BTF's string section size.
107 	 */
108 	int start_str_off;
109 
110 	/* only one of strs_data or strs_set can be non-NULL, depending on
111 	 * whether BTF is in a modifiable state (strs_set is used) or not
112 	 * (strs_data points inside raw_data)
113 	 */
114 	void *strs_data;
115 	/* a set of unique strings */
116 	struct strset *strs_set;
117 	/* whether strings are already deduplicated */
118 	bool strs_deduped;
119 
120 	/* whether base_btf should be freed in btf_free for this instance */
121 	bool owns_base;
122 
123 	/* BTF object FD, if loaded into kernel */
124 	int fd;
125 
126 	/* Pointer size (in bytes) for a target architecture of this BTF */
127 	int ptr_sz;
128 };
129 
130 static inline __u64 ptr_to_u64(const void *ptr)
131 {
132 	return (__u64) (unsigned long) ptr;
133 }
134 
135 /* Ensure given dynamically allocated memory region pointed to by *data* with
136  * capacity of *cap_cnt* elements each taking *elem_sz* bytes has enough
137  * memory to accommodate *add_cnt* new elements, assuming *cur_cnt* elements
138  * are already used. At most *max_cnt* elements can be ever allocated.
139  * If necessary, memory is reallocated and all existing data is copied over,
140  * new pointer to the memory region is stored at *data, new memory region
141  * capacity (in number of elements) is stored in *cap.
142  * On success, memory pointer to the beginning of unused memory is returned.
143  * On error, NULL is returned.
144  */
145 void *libbpf_add_mem(void **data, size_t *cap_cnt, size_t elem_sz,
146 		     size_t cur_cnt, size_t max_cnt, size_t add_cnt)
147 {
148 	size_t new_cnt;
149 	void *new_data;
150 
151 	if (cur_cnt + add_cnt <= *cap_cnt)
152 		return *data + cur_cnt * elem_sz;
153 
154 	/* requested more than the set limit */
155 	if (cur_cnt + add_cnt > max_cnt)
156 		return NULL;
157 
158 	new_cnt = *cap_cnt;
159 	new_cnt += new_cnt / 4;		  /* expand by 25% */
160 	if (new_cnt < 16)		  /* but at least 16 elements */
161 		new_cnt = 16;
162 	if (new_cnt > max_cnt)		  /* but not exceeding a set limit */
163 		new_cnt = max_cnt;
164 	if (new_cnt < cur_cnt + add_cnt)  /* also ensure we have enough memory */
165 		new_cnt = cur_cnt + add_cnt;
166 
167 	new_data = libbpf_reallocarray(*data, new_cnt, elem_sz);
168 	if (!new_data)
169 		return NULL;
170 
171 	/* zero out newly allocated portion of memory */
172 	memset(new_data + (*cap_cnt) * elem_sz, 0, (new_cnt - *cap_cnt) * elem_sz);
173 
174 	*data = new_data;
175 	*cap_cnt = new_cnt;
176 	return new_data + cur_cnt * elem_sz;
177 }
178 
179 /* Ensure given dynamically allocated memory region has enough allocated space
180  * to accommodate *need_cnt* elements of size *elem_sz* bytes each
181  */
182 int libbpf_ensure_mem(void **data, size_t *cap_cnt, size_t elem_sz, size_t need_cnt)
183 {
184 	void *p;
185 
186 	if (need_cnt <= *cap_cnt)
187 		return 0;
188 
189 	p = libbpf_add_mem(data, cap_cnt, elem_sz, *cap_cnt, SIZE_MAX, need_cnt - *cap_cnt);
190 	if (!p)
191 		return -ENOMEM;
192 
193 	return 0;
194 }
195 
196 static void *btf_add_type_offs_mem(struct btf *btf, size_t add_cnt)
197 {
198 	return libbpf_add_mem((void **)&btf->type_offs, &btf->type_offs_cap, sizeof(__u32),
199 			      btf->nr_types, BTF_MAX_NR_TYPES, add_cnt);
200 }
201 
202 static int btf_add_type_idx_entry(struct btf *btf, __u32 type_off)
203 {
204 	__u32 *p;
205 
206 	p = btf_add_type_offs_mem(btf, 1);
207 	if (!p)
208 		return -ENOMEM;
209 
210 	*p = type_off;
211 	return 0;
212 }
213 
214 static void btf_bswap_hdr(struct btf_header *h)
215 {
216 	h->magic = bswap_16(h->magic);
217 	h->hdr_len = bswap_32(h->hdr_len);
218 	h->type_off = bswap_32(h->type_off);
219 	h->type_len = bswap_32(h->type_len);
220 	h->str_off = bswap_32(h->str_off);
221 	h->str_len = bswap_32(h->str_len);
222 }
223 
224 static int btf_parse_hdr(struct btf *btf)
225 {
226 	struct btf_header *hdr = btf->hdr;
227 	__u32 meta_left;
228 
229 	if (btf->raw_size < sizeof(struct btf_header)) {
230 		pr_debug("BTF header not found\n");
231 		return -EINVAL;
232 	}
233 
234 	if (hdr->magic == bswap_16(BTF_MAGIC)) {
235 		btf->swapped_endian = true;
236 		if (bswap_32(hdr->hdr_len) != sizeof(struct btf_header)) {
237 			pr_warn("Can't load BTF with non-native endianness due to unsupported header length %u\n",
238 				bswap_32(hdr->hdr_len));
239 			return -ENOTSUP;
240 		}
241 		btf_bswap_hdr(hdr);
242 	} else if (hdr->magic != BTF_MAGIC) {
243 		pr_debug("Invalid BTF magic: %x\n", hdr->magic);
244 		return -EINVAL;
245 	}
246 
247 	if (btf->raw_size < hdr->hdr_len) {
248 		pr_debug("BTF header len %u larger than data size %u\n",
249 			 hdr->hdr_len, btf->raw_size);
250 		return -EINVAL;
251 	}
252 
253 	meta_left = btf->raw_size - hdr->hdr_len;
254 	if (meta_left < (long long)hdr->str_off + hdr->str_len) {
255 		pr_debug("Invalid BTF total size: %u\n", btf->raw_size);
256 		return -EINVAL;
257 	}
258 
259 	if ((long long)hdr->type_off + hdr->type_len > hdr->str_off) {
260 		pr_debug("Invalid BTF data sections layout: type data at %u + %u, strings data at %u + %u\n",
261 			 hdr->type_off, hdr->type_len, hdr->str_off, hdr->str_len);
262 		return -EINVAL;
263 	}
264 
265 	if (hdr->type_off % 4) {
266 		pr_debug("BTF type section is not aligned to 4 bytes\n");
267 		return -EINVAL;
268 	}
269 
270 	return 0;
271 }
272 
273 static int btf_parse_str_sec(struct btf *btf)
274 {
275 	const struct btf_header *hdr = btf->hdr;
276 	const char *start = btf->strs_data;
277 	const char *end = start + btf->hdr->str_len;
278 
279 	if (btf->base_btf && hdr->str_len == 0)
280 		return 0;
281 	if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET || end[-1]) {
282 		pr_debug("Invalid BTF string section\n");
283 		return -EINVAL;
284 	}
285 	if (!btf->base_btf && start[0]) {
286 		pr_debug("Malformed BTF string section, did you forget to provide base BTF?\n");
287 		return -EINVAL;
288 	}
289 	return 0;
290 }
291 
292 static int btf_type_size(const struct btf_type *t)
293 {
294 	const int base_size = sizeof(struct btf_type);
295 	__u16 vlen = btf_vlen(t);
296 
297 	switch (btf_kind(t)) {
298 	case BTF_KIND_FWD:
299 	case BTF_KIND_CONST:
300 	case BTF_KIND_VOLATILE:
301 	case BTF_KIND_RESTRICT:
302 	case BTF_KIND_PTR:
303 	case BTF_KIND_TYPEDEF:
304 	case BTF_KIND_FUNC:
305 	case BTF_KIND_FLOAT:
306 	case BTF_KIND_TYPE_TAG:
307 		return base_size;
308 	case BTF_KIND_INT:
309 		return base_size + sizeof(__u32);
310 	case BTF_KIND_ENUM:
311 		return base_size + vlen * sizeof(struct btf_enum);
312 	case BTF_KIND_ENUM64:
313 		return base_size + vlen * sizeof(struct btf_enum64);
314 	case BTF_KIND_ARRAY:
315 		return base_size + sizeof(struct btf_array);
316 	case BTF_KIND_STRUCT:
317 	case BTF_KIND_UNION:
318 		return base_size + vlen * sizeof(struct btf_member);
319 	case BTF_KIND_FUNC_PROTO:
320 		return base_size + vlen * sizeof(struct btf_param);
321 	case BTF_KIND_VAR:
322 		return base_size + sizeof(struct btf_var);
323 	case BTF_KIND_DATASEC:
324 		return base_size + vlen * sizeof(struct btf_var_secinfo);
325 	case BTF_KIND_DECL_TAG:
326 		return base_size + sizeof(struct btf_decl_tag);
327 	default:
328 		pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
329 		return -EINVAL;
330 	}
331 }
332 
333 static void btf_bswap_type_base(struct btf_type *t)
334 {
335 	t->name_off = bswap_32(t->name_off);
336 	t->info = bswap_32(t->info);
337 	t->type = bswap_32(t->type);
338 }
339 
340 static int btf_bswap_type_rest(struct btf_type *t)
341 {
342 	struct btf_var_secinfo *v;
343 	struct btf_enum64 *e64;
344 	struct btf_member *m;
345 	struct btf_array *a;
346 	struct btf_param *p;
347 	struct btf_enum *e;
348 	__u16 vlen = btf_vlen(t);
349 	int i;
350 
351 	switch (btf_kind(t)) {
352 	case BTF_KIND_FWD:
353 	case BTF_KIND_CONST:
354 	case BTF_KIND_VOLATILE:
355 	case BTF_KIND_RESTRICT:
356 	case BTF_KIND_PTR:
357 	case BTF_KIND_TYPEDEF:
358 	case BTF_KIND_FUNC:
359 	case BTF_KIND_FLOAT:
360 	case BTF_KIND_TYPE_TAG:
361 		return 0;
362 	case BTF_KIND_INT:
363 		*(__u32 *)(t + 1) = bswap_32(*(__u32 *)(t + 1));
364 		return 0;
365 	case BTF_KIND_ENUM:
366 		for (i = 0, e = btf_enum(t); i < vlen; i++, e++) {
367 			e->name_off = bswap_32(e->name_off);
368 			e->val = bswap_32(e->val);
369 		}
370 		return 0;
371 	case BTF_KIND_ENUM64:
372 		for (i = 0, e64 = btf_enum64(t); i < vlen; i++, e64++) {
373 			e64->name_off = bswap_32(e64->name_off);
374 			e64->val_lo32 = bswap_32(e64->val_lo32);
375 			e64->val_hi32 = bswap_32(e64->val_hi32);
376 		}
377 		return 0;
378 	case BTF_KIND_ARRAY:
379 		a = btf_array(t);
380 		a->type = bswap_32(a->type);
381 		a->index_type = bswap_32(a->index_type);
382 		a->nelems = bswap_32(a->nelems);
383 		return 0;
384 	case BTF_KIND_STRUCT:
385 	case BTF_KIND_UNION:
386 		for (i = 0, m = btf_members(t); i < vlen; i++, m++) {
387 			m->name_off = bswap_32(m->name_off);
388 			m->type = bswap_32(m->type);
389 			m->offset = bswap_32(m->offset);
390 		}
391 		return 0;
392 	case BTF_KIND_FUNC_PROTO:
393 		for (i = 0, p = btf_params(t); i < vlen; i++, p++) {
394 			p->name_off = bswap_32(p->name_off);
395 			p->type = bswap_32(p->type);
396 		}
397 		return 0;
398 	case BTF_KIND_VAR:
399 		btf_var(t)->linkage = bswap_32(btf_var(t)->linkage);
400 		return 0;
401 	case BTF_KIND_DATASEC:
402 		for (i = 0, v = btf_var_secinfos(t); i < vlen; i++, v++) {
403 			v->type = bswap_32(v->type);
404 			v->offset = bswap_32(v->offset);
405 			v->size = bswap_32(v->size);
406 		}
407 		return 0;
408 	case BTF_KIND_DECL_TAG:
409 		btf_decl_tag(t)->component_idx = bswap_32(btf_decl_tag(t)->component_idx);
410 		return 0;
411 	default:
412 		pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
413 		return -EINVAL;
414 	}
415 }
416 
417 static int btf_parse_type_sec(struct btf *btf)
418 {
419 	struct btf_header *hdr = btf->hdr;
420 	void *next_type = btf->types_data;
421 	void *end_type = next_type + hdr->type_len;
422 	int err, type_size;
423 
424 	while (next_type + sizeof(struct btf_type) <= end_type) {
425 		if (btf->swapped_endian)
426 			btf_bswap_type_base(next_type);
427 
428 		type_size = btf_type_size(next_type);
429 		if (type_size < 0)
430 			return type_size;
431 		if (next_type + type_size > end_type) {
432 			pr_warn("BTF type [%d] is malformed\n", btf->start_id + btf->nr_types);
433 			return -EINVAL;
434 		}
435 
436 		if (btf->swapped_endian && btf_bswap_type_rest(next_type))
437 			return -EINVAL;
438 
439 		err = btf_add_type_idx_entry(btf, next_type - btf->types_data);
440 		if (err)
441 			return err;
442 
443 		next_type += type_size;
444 		btf->nr_types++;
445 	}
446 
447 	if (next_type != end_type) {
448 		pr_warn("BTF types data is malformed\n");
449 		return -EINVAL;
450 	}
451 
452 	return 0;
453 }
454 
455 static int btf_validate_str(const struct btf *btf, __u32 str_off, const char *what, __u32 type_id)
456 {
457 	const char *s;
458 
459 	s = btf__str_by_offset(btf, str_off);
460 	if (!s) {
461 		pr_warn("btf: type [%u]: invalid %s (string offset %u)\n", type_id, what, str_off);
462 		return -EINVAL;
463 	}
464 
465 	return 0;
466 }
467 
468 static int btf_validate_id(const struct btf *btf, __u32 id, __u32 ctx_id)
469 {
470 	const struct btf_type *t;
471 
472 	t = btf__type_by_id(btf, id);
473 	if (!t) {
474 		pr_warn("btf: type [%u]: invalid referenced type ID %u\n", ctx_id, id);
475 		return -EINVAL;
476 	}
477 
478 	return 0;
479 }
480 
481 static int btf_validate_type(const struct btf *btf, const struct btf_type *t, __u32 id)
482 {
483 	__u32 kind = btf_kind(t);
484 	int err, i, n;
485 
486 	err = btf_validate_str(btf, t->name_off, "type name", id);
487 	if (err)
488 		return err;
489 
490 	switch (kind) {
491 	case BTF_KIND_UNKN:
492 	case BTF_KIND_INT:
493 	case BTF_KIND_FWD:
494 	case BTF_KIND_FLOAT:
495 		break;
496 	case BTF_KIND_PTR:
497 	case BTF_KIND_TYPEDEF:
498 	case BTF_KIND_VOLATILE:
499 	case BTF_KIND_CONST:
500 	case BTF_KIND_RESTRICT:
501 	case BTF_KIND_VAR:
502 	case BTF_KIND_DECL_TAG:
503 	case BTF_KIND_TYPE_TAG:
504 		err = btf_validate_id(btf, t->type, id);
505 		if (err)
506 			return err;
507 		break;
508 	case BTF_KIND_ARRAY: {
509 		const struct btf_array *a = btf_array(t);
510 
511 		err = btf_validate_id(btf, a->type, id);
512 		err = err ?: btf_validate_id(btf, a->index_type, id);
513 		if (err)
514 			return err;
515 		break;
516 	}
517 	case BTF_KIND_STRUCT:
518 	case BTF_KIND_UNION: {
519 		const struct btf_member *m = btf_members(t);
520 
521 		n = btf_vlen(t);
522 		for (i = 0; i < n; i++, m++) {
523 			err = btf_validate_str(btf, m->name_off, "field name", id);
524 			err = err ?: btf_validate_id(btf, m->type, id);
525 			if (err)
526 				return err;
527 		}
528 		break;
529 	}
530 	case BTF_KIND_ENUM: {
531 		const struct btf_enum *m = btf_enum(t);
532 
533 		n = btf_vlen(t);
534 		for (i = 0; i < n; i++, m++) {
535 			err = btf_validate_str(btf, m->name_off, "enum name", id);
536 			if (err)
537 				return err;
538 		}
539 		break;
540 	}
541 	case BTF_KIND_ENUM64: {
542 		const struct btf_enum64 *m = btf_enum64(t);
543 
544 		n = btf_vlen(t);
545 		for (i = 0; i < n; i++, m++) {
546 			err = btf_validate_str(btf, m->name_off, "enum name", id);
547 			if (err)
548 				return err;
549 		}
550 		break;
551 	}
552 	case BTF_KIND_FUNC: {
553 		const struct btf_type *ft;
554 
555 		err = btf_validate_id(btf, t->type, id);
556 		if (err)
557 			return err;
558 		ft = btf__type_by_id(btf, t->type);
559 		if (btf_kind(ft) != BTF_KIND_FUNC_PROTO) {
560 			pr_warn("btf: type [%u]: referenced type [%u] is not FUNC_PROTO\n", id, t->type);
561 			return -EINVAL;
562 		}
563 		break;
564 	}
565 	case BTF_KIND_FUNC_PROTO: {
566 		const struct btf_param *m = btf_params(t);
567 
568 		n = btf_vlen(t);
569 		for (i = 0; i < n; i++, m++) {
570 			err = btf_validate_str(btf, m->name_off, "param name", id);
571 			err = err ?: btf_validate_id(btf, m->type, id);
572 			if (err)
573 				return err;
574 		}
575 		break;
576 	}
577 	case BTF_KIND_DATASEC: {
578 		const struct btf_var_secinfo *m = btf_var_secinfos(t);
579 
580 		n = btf_vlen(t);
581 		for (i = 0; i < n; i++, m++) {
582 			err = btf_validate_id(btf, m->type, id);
583 			if (err)
584 				return err;
585 		}
586 		break;
587 	}
588 	default:
589 		pr_warn("btf: type [%u]: unrecognized kind %u\n", id, kind);
590 		return -EINVAL;
591 	}
592 	return 0;
593 }
594 
595 /* Validate basic sanity of BTF. It's intentionally less thorough than
596  * kernel's validation and validates only properties of BTF that libbpf relies
597  * on to be correct (e.g., valid type IDs, valid string offsets, etc)
598  */
599 static int btf_sanity_check(const struct btf *btf)
600 {
601 	const struct btf_type *t;
602 	__u32 i, n = btf__type_cnt(btf);
603 	int err;
604 
605 	for (i = btf->start_id; i < n; i++) {
606 		t = btf_type_by_id(btf, i);
607 		err = btf_validate_type(btf, t, i);
608 		if (err)
609 			return err;
610 	}
611 	return 0;
612 }
613 
614 __u32 btf__type_cnt(const struct btf *btf)
615 {
616 	return btf->start_id + btf->nr_types;
617 }
618 
619 const struct btf *btf__base_btf(const struct btf *btf)
620 {
621 	return btf->base_btf;
622 }
623 
624 /* internal helper returning non-const pointer to a type */
625 struct btf_type *btf_type_by_id(const struct btf *btf, __u32 type_id)
626 {
627 	if (type_id == 0)
628 		return &btf_void;
629 	if (type_id < btf->start_id)
630 		return btf_type_by_id(btf->base_btf, type_id);
631 	return btf->types_data + btf->type_offs[type_id - btf->start_id];
632 }
633 
634 const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
635 {
636 	if (type_id >= btf->start_id + btf->nr_types)
637 		return errno = EINVAL, NULL;
638 	return btf_type_by_id((struct btf *)btf, type_id);
639 }
640 
641 static int determine_ptr_size(const struct btf *btf)
642 {
643 	static const char * const long_aliases[] = {
644 		"long",
645 		"long int",
646 		"int long",
647 		"unsigned long",
648 		"long unsigned",
649 		"unsigned long int",
650 		"unsigned int long",
651 		"long unsigned int",
652 		"long int unsigned",
653 		"int unsigned long",
654 		"int long unsigned",
655 	};
656 	const struct btf_type *t;
657 	const char *name;
658 	int i, j, n;
659 
660 	if (btf->base_btf && btf->base_btf->ptr_sz > 0)
661 		return btf->base_btf->ptr_sz;
662 
663 	n = btf__type_cnt(btf);
664 	for (i = 1; i < n; i++) {
665 		t = btf__type_by_id(btf, i);
666 		if (!btf_is_int(t))
667 			continue;
668 
669 		if (t->size != 4 && t->size != 8)
670 			continue;
671 
672 		name = btf__name_by_offset(btf, t->name_off);
673 		if (!name)
674 			continue;
675 
676 		for (j = 0; j < ARRAY_SIZE(long_aliases); j++) {
677 			if (strcmp(name, long_aliases[j]) == 0)
678 				return t->size;
679 		}
680 	}
681 
682 	return -1;
683 }
684 
685 static size_t btf_ptr_sz(const struct btf *btf)
686 {
687 	if (!btf->ptr_sz)
688 		((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
689 	return btf->ptr_sz < 0 ? sizeof(void *) : btf->ptr_sz;
690 }
691 
692 /* Return pointer size this BTF instance assumes. The size is heuristically
693  * determined by looking for 'long' or 'unsigned long' integer type and
694  * recording its size in bytes. If BTF type information doesn't have any such
695  * type, this function returns 0. In the latter case, native architecture's
696  * pointer size is assumed, so will be either 4 or 8, depending on
697  * architecture that libbpf was compiled for. It's possible to override
698  * guessed value by using btf__set_pointer_size() API.
699  */
700 size_t btf__pointer_size(const struct btf *btf)
701 {
702 	if (!btf->ptr_sz)
703 		((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
704 
705 	if (btf->ptr_sz < 0)
706 		/* not enough BTF type info to guess */
707 		return 0;
708 
709 	return btf->ptr_sz;
710 }
711 
712 /* Override or set pointer size in bytes. Only values of 4 and 8 are
713  * supported.
714  */
715 int btf__set_pointer_size(struct btf *btf, size_t ptr_sz)
716 {
717 	if (ptr_sz != 4 && ptr_sz != 8)
718 		return libbpf_err(-EINVAL);
719 	btf->ptr_sz = ptr_sz;
720 	return 0;
721 }
722 
723 static bool is_host_big_endian(void)
724 {
725 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
726 	return false;
727 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
728 	return true;
729 #else
730 # error "Unrecognized __BYTE_ORDER__"
731 #endif
732 }
733 
734 enum btf_endianness btf__endianness(const struct btf *btf)
735 {
736 	if (is_host_big_endian())
737 		return btf->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
738 	else
739 		return btf->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
740 }
741 
742 int btf__set_endianness(struct btf *btf, enum btf_endianness endian)
743 {
744 	if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
745 		return libbpf_err(-EINVAL);
746 
747 	btf->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
748 	if (!btf->swapped_endian) {
749 		free(btf->raw_data_swapped);
750 		btf->raw_data_swapped = NULL;
751 	}
752 	return 0;
753 }
754 
755 static bool btf_type_is_void(const struct btf_type *t)
756 {
757 	return t == &btf_void || btf_is_fwd(t);
758 }
759 
760 static bool btf_type_is_void_or_null(const struct btf_type *t)
761 {
762 	return !t || btf_type_is_void(t);
763 }
764 
765 #define MAX_RESOLVE_DEPTH 32
766 
767 __s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
768 {
769 	const struct btf_array *array;
770 	const struct btf_type *t;
771 	__u32 nelems = 1;
772 	__s64 size = -1;
773 	int i;
774 
775 	t = btf__type_by_id(btf, type_id);
776 	for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t); i++) {
777 		switch (btf_kind(t)) {
778 		case BTF_KIND_INT:
779 		case BTF_KIND_STRUCT:
780 		case BTF_KIND_UNION:
781 		case BTF_KIND_ENUM:
782 		case BTF_KIND_ENUM64:
783 		case BTF_KIND_DATASEC:
784 		case BTF_KIND_FLOAT:
785 			size = t->size;
786 			goto done;
787 		case BTF_KIND_PTR:
788 			size = btf_ptr_sz(btf);
789 			goto done;
790 		case BTF_KIND_TYPEDEF:
791 		case BTF_KIND_VOLATILE:
792 		case BTF_KIND_CONST:
793 		case BTF_KIND_RESTRICT:
794 		case BTF_KIND_VAR:
795 		case BTF_KIND_DECL_TAG:
796 		case BTF_KIND_TYPE_TAG:
797 			type_id = t->type;
798 			break;
799 		case BTF_KIND_ARRAY:
800 			array = btf_array(t);
801 			if (nelems && array->nelems > UINT32_MAX / nelems)
802 				return libbpf_err(-E2BIG);
803 			nelems *= array->nelems;
804 			type_id = array->type;
805 			break;
806 		default:
807 			return libbpf_err(-EINVAL);
808 		}
809 
810 		t = btf__type_by_id(btf, type_id);
811 	}
812 
813 done:
814 	if (size < 0)
815 		return libbpf_err(-EINVAL);
816 	if (nelems && size > UINT32_MAX / nelems)
817 		return libbpf_err(-E2BIG);
818 
819 	return nelems * size;
820 }
821 
822 int btf__align_of(const struct btf *btf, __u32 id)
823 {
824 	const struct btf_type *t = btf__type_by_id(btf, id);
825 	__u16 kind = btf_kind(t);
826 
827 	switch (kind) {
828 	case BTF_KIND_INT:
829 	case BTF_KIND_ENUM:
830 	case BTF_KIND_ENUM64:
831 	case BTF_KIND_FLOAT:
832 		return min(btf_ptr_sz(btf), (size_t)t->size);
833 	case BTF_KIND_PTR:
834 		return btf_ptr_sz(btf);
835 	case BTF_KIND_TYPEDEF:
836 	case BTF_KIND_VOLATILE:
837 	case BTF_KIND_CONST:
838 	case BTF_KIND_RESTRICT:
839 	case BTF_KIND_TYPE_TAG:
840 		return btf__align_of(btf, t->type);
841 	case BTF_KIND_ARRAY:
842 		return btf__align_of(btf, btf_array(t)->type);
843 	case BTF_KIND_STRUCT:
844 	case BTF_KIND_UNION: {
845 		const struct btf_member *m = btf_members(t);
846 		__u16 vlen = btf_vlen(t);
847 		int i, max_align = 1, align;
848 
849 		for (i = 0; i < vlen; i++, m++) {
850 			align = btf__align_of(btf, m->type);
851 			if (align <= 0)
852 				return libbpf_err(align);
853 			max_align = max(max_align, align);
854 
855 			/* if field offset isn't aligned according to field
856 			 * type's alignment, then struct must be packed
857 			 */
858 			if (btf_member_bitfield_size(t, i) == 0 &&
859 			    (m->offset % (8 * align)) != 0)
860 				return 1;
861 		}
862 
863 		/* if struct/union size isn't a multiple of its alignment,
864 		 * then struct must be packed
865 		 */
866 		if ((t->size % max_align) != 0)
867 			return 1;
868 
869 		return max_align;
870 	}
871 	default:
872 		pr_warn("unsupported BTF_KIND:%u\n", btf_kind(t));
873 		return errno = EINVAL, 0;
874 	}
875 }
876 
877 int btf__resolve_type(const struct btf *btf, __u32 type_id)
878 {
879 	const struct btf_type *t;
880 	int depth = 0;
881 
882 	t = btf__type_by_id(btf, type_id);
883 	while (depth < MAX_RESOLVE_DEPTH &&
884 	       !btf_type_is_void_or_null(t) &&
885 	       (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
886 		type_id = t->type;
887 		t = btf__type_by_id(btf, type_id);
888 		depth++;
889 	}
890 
891 	if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
892 		return libbpf_err(-EINVAL);
893 
894 	return type_id;
895 }
896 
897 __s32 btf__find_by_name(const struct btf *btf, const char *type_name)
898 {
899 	__u32 i, nr_types = btf__type_cnt(btf);
900 
901 	if (!strcmp(type_name, "void"))
902 		return 0;
903 
904 	for (i = 1; i < nr_types; i++) {
905 		const struct btf_type *t = btf__type_by_id(btf, i);
906 		const char *name = btf__name_by_offset(btf, t->name_off);
907 
908 		if (name && !strcmp(type_name, name))
909 			return i;
910 	}
911 
912 	return libbpf_err(-ENOENT);
913 }
914 
915 static __s32 btf_find_by_name_kind(const struct btf *btf, int start_id,
916 				   const char *type_name, __u32 kind)
917 {
918 	__u32 i, nr_types = btf__type_cnt(btf);
919 
920 	if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
921 		return 0;
922 
923 	for (i = start_id; i < nr_types; i++) {
924 		const struct btf_type *t = btf__type_by_id(btf, i);
925 		const char *name;
926 
927 		if (btf_kind(t) != kind)
928 			continue;
929 		name = btf__name_by_offset(btf, t->name_off);
930 		if (name && !strcmp(type_name, name))
931 			return i;
932 	}
933 
934 	return libbpf_err(-ENOENT);
935 }
936 
937 __s32 btf__find_by_name_kind_own(const struct btf *btf, const char *type_name,
938 				 __u32 kind)
939 {
940 	return btf_find_by_name_kind(btf, btf->start_id, type_name, kind);
941 }
942 
943 __s32 btf__find_by_name_kind(const struct btf *btf, const char *type_name,
944 			     __u32 kind)
945 {
946 	return btf_find_by_name_kind(btf, 1, type_name, kind);
947 }
948 
949 static bool btf_is_modifiable(const struct btf *btf)
950 {
951 	return (void *)btf->hdr != btf->raw_data;
952 }
953 
954 void btf__free(struct btf *btf)
955 {
956 	if (IS_ERR_OR_NULL(btf))
957 		return;
958 
959 	if (btf->fd >= 0)
960 		close(btf->fd);
961 
962 	if (btf_is_modifiable(btf)) {
963 		/* if BTF was modified after loading, it will have a split
964 		 * in-memory representation for header, types, and strings
965 		 * sections, so we need to free all of them individually. It
966 		 * might still have a cached contiguous raw data present,
967 		 * which will be unconditionally freed below.
968 		 */
969 		free(btf->hdr);
970 		free(btf->types_data);
971 		strset__free(btf->strs_set);
972 	}
973 	free(btf->raw_data);
974 	free(btf->raw_data_swapped);
975 	free(btf->type_offs);
976 	if (btf->owns_base)
977 		btf__free(btf->base_btf);
978 	free(btf);
979 }
980 
981 static struct btf *btf_new_empty(struct btf *base_btf)
982 {
983 	struct btf *btf;
984 
985 	btf = calloc(1, sizeof(*btf));
986 	if (!btf)
987 		return ERR_PTR(-ENOMEM);
988 
989 	btf->nr_types = 0;
990 	btf->start_id = 1;
991 	btf->start_str_off = 0;
992 	btf->fd = -1;
993 	btf->ptr_sz = sizeof(void *);
994 	btf->swapped_endian = false;
995 
996 	if (base_btf) {
997 		btf->base_btf = base_btf;
998 		btf->start_id = btf__type_cnt(base_btf);
999 		btf->start_str_off = base_btf->hdr->str_len;
1000 		btf->swapped_endian = base_btf->swapped_endian;
1001 	}
1002 
1003 	/* +1 for empty string at offset 0 */
1004 	btf->raw_size = sizeof(struct btf_header) + (base_btf ? 0 : 1);
1005 	btf->raw_data = calloc(1, btf->raw_size);
1006 	if (!btf->raw_data) {
1007 		free(btf);
1008 		return ERR_PTR(-ENOMEM);
1009 	}
1010 
1011 	btf->hdr = btf->raw_data;
1012 	btf->hdr->hdr_len = sizeof(struct btf_header);
1013 	btf->hdr->magic = BTF_MAGIC;
1014 	btf->hdr->version = BTF_VERSION;
1015 
1016 	btf->types_data = btf->raw_data + btf->hdr->hdr_len;
1017 	btf->strs_data = btf->raw_data + btf->hdr->hdr_len;
1018 	btf->hdr->str_len = base_btf ? 0 : 1; /* empty string at offset 0 */
1019 
1020 	return btf;
1021 }
1022 
1023 struct btf *btf__new_empty(void)
1024 {
1025 	return libbpf_ptr(btf_new_empty(NULL));
1026 }
1027 
1028 struct btf *btf__new_empty_split(struct btf *base_btf)
1029 {
1030 	return libbpf_ptr(btf_new_empty(base_btf));
1031 }
1032 
1033 static struct btf *btf_new(const void *data, __u32 size, struct btf *base_btf)
1034 {
1035 	struct btf *btf;
1036 	int err;
1037 
1038 	btf = calloc(1, sizeof(struct btf));
1039 	if (!btf)
1040 		return ERR_PTR(-ENOMEM);
1041 
1042 	btf->nr_types = 0;
1043 	btf->start_id = 1;
1044 	btf->start_str_off = 0;
1045 	btf->fd = -1;
1046 
1047 	if (base_btf) {
1048 		btf->base_btf = base_btf;
1049 		btf->start_id = btf__type_cnt(base_btf);
1050 		btf->start_str_off = base_btf->hdr->str_len;
1051 	}
1052 
1053 	btf->raw_data = malloc(size);
1054 	if (!btf->raw_data) {
1055 		err = -ENOMEM;
1056 		goto done;
1057 	}
1058 	memcpy(btf->raw_data, data, size);
1059 	btf->raw_size = size;
1060 
1061 	btf->hdr = btf->raw_data;
1062 	err = btf_parse_hdr(btf);
1063 	if (err)
1064 		goto done;
1065 
1066 	btf->strs_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->str_off;
1067 	btf->types_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->type_off;
1068 
1069 	err = btf_parse_str_sec(btf);
1070 	err = err ?: btf_parse_type_sec(btf);
1071 	err = err ?: btf_sanity_check(btf);
1072 	if (err)
1073 		goto done;
1074 
1075 done:
1076 	if (err) {
1077 		btf__free(btf);
1078 		return ERR_PTR(err);
1079 	}
1080 
1081 	return btf;
1082 }
1083 
1084 struct btf *btf__new(const void *data, __u32 size)
1085 {
1086 	return libbpf_ptr(btf_new(data, size, NULL));
1087 }
1088 
1089 struct btf *btf__new_split(const void *data, __u32 size, struct btf *base_btf)
1090 {
1091 	return libbpf_ptr(btf_new(data, size, base_btf));
1092 }
1093 
1094 struct btf_elf_secs {
1095 	Elf_Data *btf_data;
1096 	Elf_Data *btf_ext_data;
1097 	Elf_Data *btf_base_data;
1098 };
1099 
1100 static int btf_find_elf_sections(Elf *elf, const char *path, struct btf_elf_secs *secs)
1101 {
1102 	Elf_Scn *scn = NULL;
1103 	Elf_Data *data;
1104 	GElf_Ehdr ehdr;
1105 	size_t shstrndx;
1106 	int idx = 0;
1107 
1108 	if (!gelf_getehdr(elf, &ehdr)) {
1109 		pr_warn("failed to get EHDR from %s\n", path);
1110 		goto err;
1111 	}
1112 
1113 	if (elf_getshdrstrndx(elf, &shstrndx)) {
1114 		pr_warn("failed to get section names section index for %s\n",
1115 			path);
1116 		goto err;
1117 	}
1118 
1119 	if (!elf_rawdata(elf_getscn(elf, shstrndx), NULL)) {
1120 		pr_warn("failed to get e_shstrndx from %s\n", path);
1121 		goto err;
1122 	}
1123 
1124 	while ((scn = elf_nextscn(elf, scn)) != NULL) {
1125 		Elf_Data **field;
1126 		GElf_Shdr sh;
1127 		char *name;
1128 
1129 		idx++;
1130 		if (gelf_getshdr(scn, &sh) != &sh) {
1131 			pr_warn("failed to get section(%d) header from %s\n",
1132 				idx, path);
1133 			goto err;
1134 		}
1135 		name = elf_strptr(elf, shstrndx, sh.sh_name);
1136 		if (!name) {
1137 			pr_warn("failed to get section(%d) name from %s\n",
1138 				idx, path);
1139 			goto err;
1140 		}
1141 
1142 		if (strcmp(name, BTF_ELF_SEC) == 0)
1143 			field = &secs->btf_data;
1144 		else if (strcmp(name, BTF_EXT_ELF_SEC) == 0)
1145 			field = &secs->btf_ext_data;
1146 		else if (strcmp(name, BTF_BASE_ELF_SEC) == 0)
1147 			field = &secs->btf_base_data;
1148 		else
1149 			continue;
1150 
1151 		data = elf_getdata(scn, 0);
1152 		if (!data) {
1153 			pr_warn("failed to get section(%d, %s) data from %s\n",
1154 				idx, name, path);
1155 			goto err;
1156 		}
1157 		*field = data;
1158 	}
1159 
1160 	return 0;
1161 
1162 err:
1163 	return -LIBBPF_ERRNO__FORMAT;
1164 }
1165 
1166 static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,
1167 				 struct btf_ext **btf_ext)
1168 {
1169 	struct btf_elf_secs secs = {};
1170 	struct btf *dist_base_btf = NULL;
1171 	struct btf *btf = NULL;
1172 	int err = 0, fd = -1;
1173 	Elf *elf = NULL;
1174 
1175 	if (elf_version(EV_CURRENT) == EV_NONE) {
1176 		pr_warn("failed to init libelf for %s\n", path);
1177 		return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1178 	}
1179 
1180 	fd = open(path, O_RDONLY | O_CLOEXEC);
1181 	if (fd < 0) {
1182 		err = -errno;
1183 		pr_warn("failed to open %s: %s\n", path, errstr(err));
1184 		return ERR_PTR(err);
1185 	}
1186 
1187 	elf = elf_begin(fd, ELF_C_READ, NULL);
1188 	if (!elf) {
1189 		err = -LIBBPF_ERRNO__FORMAT;
1190 		pr_warn("failed to open %s as ELF file\n", path);
1191 		goto done;
1192 	}
1193 
1194 	err = btf_find_elf_sections(elf, path, &secs);
1195 	if (err)
1196 		goto done;
1197 
1198 	if (!secs.btf_data) {
1199 		pr_warn("failed to find '%s' ELF section in %s\n", BTF_ELF_SEC, path);
1200 		err = -ENODATA;
1201 		goto done;
1202 	}
1203 
1204 	if (secs.btf_base_data) {
1205 		dist_base_btf = btf_new(secs.btf_base_data->d_buf, secs.btf_base_data->d_size,
1206 					NULL);
1207 		if (IS_ERR(dist_base_btf)) {
1208 			err = PTR_ERR(dist_base_btf);
1209 			dist_base_btf = NULL;
1210 			goto done;
1211 		}
1212 	}
1213 
1214 	btf = btf_new(secs.btf_data->d_buf, secs.btf_data->d_size,
1215 		      dist_base_btf ?: base_btf);
1216 	if (IS_ERR(btf)) {
1217 		err = PTR_ERR(btf);
1218 		goto done;
1219 	}
1220 	if (dist_base_btf && base_btf) {
1221 		err = btf__relocate(btf, base_btf);
1222 		if (err)
1223 			goto done;
1224 		btf__free(dist_base_btf);
1225 		dist_base_btf = NULL;
1226 	}
1227 
1228 	if (dist_base_btf)
1229 		btf->owns_base = true;
1230 
1231 	switch (gelf_getclass(elf)) {
1232 	case ELFCLASS32:
1233 		btf__set_pointer_size(btf, 4);
1234 		break;
1235 	case ELFCLASS64:
1236 		btf__set_pointer_size(btf, 8);
1237 		break;
1238 	default:
1239 		pr_warn("failed to get ELF class (bitness) for %s\n", path);
1240 		break;
1241 	}
1242 
1243 	if (btf_ext && secs.btf_ext_data) {
1244 		*btf_ext = btf_ext__new(secs.btf_ext_data->d_buf, secs.btf_ext_data->d_size);
1245 		if (IS_ERR(*btf_ext)) {
1246 			err = PTR_ERR(*btf_ext);
1247 			goto done;
1248 		}
1249 	} else if (btf_ext) {
1250 		*btf_ext = NULL;
1251 	}
1252 done:
1253 	if (elf)
1254 		elf_end(elf);
1255 	close(fd);
1256 
1257 	if (!err)
1258 		return btf;
1259 
1260 	if (btf_ext)
1261 		btf_ext__free(*btf_ext);
1262 	btf__free(dist_base_btf);
1263 	btf__free(btf);
1264 
1265 	return ERR_PTR(err);
1266 }
1267 
1268 struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
1269 {
1270 	return libbpf_ptr(btf_parse_elf(path, NULL, btf_ext));
1271 }
1272 
1273 struct btf *btf__parse_elf_split(const char *path, struct btf *base_btf)
1274 {
1275 	return libbpf_ptr(btf_parse_elf(path, base_btf, NULL));
1276 }
1277 
1278 static struct btf *btf_parse_raw(const char *path, struct btf *base_btf)
1279 {
1280 	struct btf *btf = NULL;
1281 	void *data = NULL;
1282 	FILE *f = NULL;
1283 	__u16 magic;
1284 	int err = 0;
1285 	long sz;
1286 
1287 	f = fopen(path, "rbe");
1288 	if (!f) {
1289 		err = -errno;
1290 		goto err_out;
1291 	}
1292 
1293 	/* check BTF magic */
1294 	if (fread(&magic, 1, sizeof(magic), f) < sizeof(magic)) {
1295 		err = -EIO;
1296 		goto err_out;
1297 	}
1298 	if (magic != BTF_MAGIC && magic != bswap_16(BTF_MAGIC)) {
1299 		/* definitely not a raw BTF */
1300 		err = -EPROTO;
1301 		goto err_out;
1302 	}
1303 
1304 	/* get file size */
1305 	if (fseek(f, 0, SEEK_END)) {
1306 		err = -errno;
1307 		goto err_out;
1308 	}
1309 	sz = ftell(f);
1310 	if (sz < 0) {
1311 		err = -errno;
1312 		goto err_out;
1313 	}
1314 	/* rewind to the start */
1315 	if (fseek(f, 0, SEEK_SET)) {
1316 		err = -errno;
1317 		goto err_out;
1318 	}
1319 
1320 	/* pre-alloc memory and read all of BTF data */
1321 	data = malloc(sz);
1322 	if (!data) {
1323 		err = -ENOMEM;
1324 		goto err_out;
1325 	}
1326 	if (fread(data, 1, sz, f) < sz) {
1327 		err = -EIO;
1328 		goto err_out;
1329 	}
1330 
1331 	/* finally parse BTF data */
1332 	btf = btf_new(data, sz, base_btf);
1333 
1334 err_out:
1335 	free(data);
1336 	if (f)
1337 		fclose(f);
1338 	return err ? ERR_PTR(err) : btf;
1339 }
1340 
1341 struct btf *btf__parse_raw(const char *path)
1342 {
1343 	return libbpf_ptr(btf_parse_raw(path, NULL));
1344 }
1345 
1346 struct btf *btf__parse_raw_split(const char *path, struct btf *base_btf)
1347 {
1348 	return libbpf_ptr(btf_parse_raw(path, base_btf));
1349 }
1350 
1351 static struct btf *btf_parse(const char *path, struct btf *base_btf, struct btf_ext **btf_ext)
1352 {
1353 	struct btf *btf;
1354 	int err;
1355 
1356 	if (btf_ext)
1357 		*btf_ext = NULL;
1358 
1359 	btf = btf_parse_raw(path, base_btf);
1360 	err = libbpf_get_error(btf);
1361 	if (!err)
1362 		return btf;
1363 	if (err != -EPROTO)
1364 		return ERR_PTR(err);
1365 	return btf_parse_elf(path, base_btf, btf_ext);
1366 }
1367 
1368 struct btf *btf__parse(const char *path, struct btf_ext **btf_ext)
1369 {
1370 	return libbpf_ptr(btf_parse(path, NULL, btf_ext));
1371 }
1372 
1373 struct btf *btf__parse_split(const char *path, struct btf *base_btf)
1374 {
1375 	return libbpf_ptr(btf_parse(path, base_btf, NULL));
1376 }
1377 
1378 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian);
1379 
1380 int btf_load_into_kernel(struct btf *btf,
1381 			 char *log_buf, size_t log_sz, __u32 log_level,
1382 			 int token_fd)
1383 {
1384 	LIBBPF_OPTS(bpf_btf_load_opts, opts);
1385 	__u32 buf_sz = 0, raw_size;
1386 	char *buf = NULL, *tmp;
1387 	void *raw_data;
1388 	int err = 0;
1389 
1390 	if (btf->fd >= 0)
1391 		return libbpf_err(-EEXIST);
1392 	if (log_sz && !log_buf)
1393 		return libbpf_err(-EINVAL);
1394 
1395 	/* cache native raw data representation */
1396 	raw_data = btf_get_raw_data(btf, &raw_size, false);
1397 	if (!raw_data) {
1398 		err = -ENOMEM;
1399 		goto done;
1400 	}
1401 	btf->raw_size = raw_size;
1402 	btf->raw_data = raw_data;
1403 
1404 retry_load:
1405 	/* if log_level is 0, we won't provide log_buf/log_size to the kernel,
1406 	 * initially. Only if BTF loading fails, we bump log_level to 1 and
1407 	 * retry, using either auto-allocated or custom log_buf. This way
1408 	 * non-NULL custom log_buf provides a buffer just in case, but hopes
1409 	 * for successful load and no need for log_buf.
1410 	 */
1411 	if (log_level) {
1412 		/* if caller didn't provide custom log_buf, we'll keep
1413 		 * allocating our own progressively bigger buffers for BTF
1414 		 * verification log
1415 		 */
1416 		if (!log_buf) {
1417 			buf_sz = max((__u32)BPF_LOG_BUF_SIZE, buf_sz * 2);
1418 			tmp = realloc(buf, buf_sz);
1419 			if (!tmp) {
1420 				err = -ENOMEM;
1421 				goto done;
1422 			}
1423 			buf = tmp;
1424 			buf[0] = '\0';
1425 		}
1426 
1427 		opts.log_buf = log_buf ? log_buf : buf;
1428 		opts.log_size = log_buf ? log_sz : buf_sz;
1429 		opts.log_level = log_level;
1430 	}
1431 
1432 	opts.token_fd = token_fd;
1433 	if (token_fd)
1434 		opts.btf_flags |= BPF_F_TOKEN_FD;
1435 
1436 	btf->fd = bpf_btf_load(raw_data, raw_size, &opts);
1437 	if (btf->fd < 0) {
1438 		/* time to turn on verbose mode and try again */
1439 		if (log_level == 0) {
1440 			log_level = 1;
1441 			goto retry_load;
1442 		}
1443 		/* only retry if caller didn't provide custom log_buf, but
1444 		 * make sure we can never overflow buf_sz
1445 		 */
1446 		if (!log_buf && errno == ENOSPC && buf_sz <= UINT_MAX / 2)
1447 			goto retry_load;
1448 
1449 		err = -errno;
1450 		pr_warn("BTF loading error: %s\n", errstr(err));
1451 		/* don't print out contents of custom log_buf */
1452 		if (!log_buf && buf[0])
1453 			pr_warn("-- BEGIN BTF LOAD LOG ---\n%s\n-- END BTF LOAD LOG --\n", buf);
1454 	}
1455 
1456 done:
1457 	free(buf);
1458 	return libbpf_err(err);
1459 }
1460 
1461 int btf__load_into_kernel(struct btf *btf)
1462 {
1463 	return btf_load_into_kernel(btf, NULL, 0, 0, 0);
1464 }
1465 
1466 int btf__fd(const struct btf *btf)
1467 {
1468 	return btf->fd;
1469 }
1470 
1471 void btf__set_fd(struct btf *btf, int fd)
1472 {
1473 	btf->fd = fd;
1474 }
1475 
1476 static const void *btf_strs_data(const struct btf *btf)
1477 {
1478 	return btf->strs_data ? btf->strs_data : strset__data(btf->strs_set);
1479 }
1480 
1481 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian)
1482 {
1483 	struct btf_header *hdr = btf->hdr;
1484 	struct btf_type *t;
1485 	void *data, *p;
1486 	__u32 data_sz;
1487 	int i;
1488 
1489 	data = swap_endian ? btf->raw_data_swapped : btf->raw_data;
1490 	if (data) {
1491 		*size = btf->raw_size;
1492 		return data;
1493 	}
1494 
1495 	data_sz = hdr->hdr_len + hdr->type_len + hdr->str_len;
1496 	data = calloc(1, data_sz);
1497 	if (!data)
1498 		return NULL;
1499 	p = data;
1500 
1501 	memcpy(p, hdr, hdr->hdr_len);
1502 	if (swap_endian)
1503 		btf_bswap_hdr(p);
1504 	p += hdr->hdr_len;
1505 
1506 	memcpy(p, btf->types_data, hdr->type_len);
1507 	if (swap_endian) {
1508 		for (i = 0; i < btf->nr_types; i++) {
1509 			t = p + btf->type_offs[i];
1510 			/* btf_bswap_type_rest() relies on native t->info, so
1511 			 * we swap base type info after we swapped all the
1512 			 * additional information
1513 			 */
1514 			if (btf_bswap_type_rest(t))
1515 				goto err_out;
1516 			btf_bswap_type_base(t);
1517 		}
1518 	}
1519 	p += hdr->type_len;
1520 
1521 	memcpy(p, btf_strs_data(btf), hdr->str_len);
1522 	p += hdr->str_len;
1523 
1524 	*size = data_sz;
1525 	return data;
1526 err_out:
1527 	free(data);
1528 	return NULL;
1529 }
1530 
1531 const void *btf__raw_data(const struct btf *btf_ro, __u32 *size)
1532 {
1533 	struct btf *btf = (struct btf *)btf_ro;
1534 	__u32 data_sz;
1535 	void *data;
1536 
1537 	data = btf_get_raw_data(btf, &data_sz, btf->swapped_endian);
1538 	if (!data)
1539 		return errno = ENOMEM, NULL;
1540 
1541 	btf->raw_size = data_sz;
1542 	if (btf->swapped_endian)
1543 		btf->raw_data_swapped = data;
1544 	else
1545 		btf->raw_data = data;
1546 	*size = data_sz;
1547 	return data;
1548 }
1549 
1550 __attribute__((alias("btf__raw_data")))
1551 const void *btf__get_raw_data(const struct btf *btf, __u32 *size);
1552 
1553 const char *btf__str_by_offset(const struct btf *btf, __u32 offset)
1554 {
1555 	if (offset < btf->start_str_off)
1556 		return btf__str_by_offset(btf->base_btf, offset);
1557 	else if (offset - btf->start_str_off < btf->hdr->str_len)
1558 		return btf_strs_data(btf) + (offset - btf->start_str_off);
1559 	else
1560 		return errno = EINVAL, NULL;
1561 }
1562 
1563 const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
1564 {
1565 	return btf__str_by_offset(btf, offset);
1566 }
1567 
1568 struct btf *btf_get_from_fd(int btf_fd, struct btf *base_btf)
1569 {
1570 	struct bpf_btf_info btf_info;
1571 	__u32 len = sizeof(btf_info);
1572 	__u32 last_size;
1573 	struct btf *btf;
1574 	void *ptr;
1575 	int err;
1576 
1577 	/* we won't know btf_size until we call bpf_btf_get_info_by_fd(). so
1578 	 * let's start with a sane default - 4KiB here - and resize it only if
1579 	 * bpf_btf_get_info_by_fd() needs a bigger buffer.
1580 	 */
1581 	last_size = 4096;
1582 	ptr = malloc(last_size);
1583 	if (!ptr)
1584 		return ERR_PTR(-ENOMEM);
1585 
1586 	memset(&btf_info, 0, sizeof(btf_info));
1587 	btf_info.btf = ptr_to_u64(ptr);
1588 	btf_info.btf_size = last_size;
1589 	err = bpf_btf_get_info_by_fd(btf_fd, &btf_info, &len);
1590 
1591 	if (!err && btf_info.btf_size > last_size) {
1592 		void *temp_ptr;
1593 
1594 		last_size = btf_info.btf_size;
1595 		temp_ptr = realloc(ptr, last_size);
1596 		if (!temp_ptr) {
1597 			btf = ERR_PTR(-ENOMEM);
1598 			goto exit_free;
1599 		}
1600 		ptr = temp_ptr;
1601 
1602 		len = sizeof(btf_info);
1603 		memset(&btf_info, 0, sizeof(btf_info));
1604 		btf_info.btf = ptr_to_u64(ptr);
1605 		btf_info.btf_size = last_size;
1606 
1607 		err = bpf_btf_get_info_by_fd(btf_fd, &btf_info, &len);
1608 	}
1609 
1610 	if (err || btf_info.btf_size > last_size) {
1611 		btf = err ? ERR_PTR(-errno) : ERR_PTR(-E2BIG);
1612 		goto exit_free;
1613 	}
1614 
1615 	btf = btf_new(ptr, btf_info.btf_size, base_btf);
1616 
1617 exit_free:
1618 	free(ptr);
1619 	return btf;
1620 }
1621 
1622 struct btf *btf__load_from_kernel_by_id_split(__u32 id, struct btf *base_btf)
1623 {
1624 	struct btf *btf;
1625 	int btf_fd;
1626 
1627 	btf_fd = bpf_btf_get_fd_by_id(id);
1628 	if (btf_fd < 0)
1629 		return libbpf_err_ptr(-errno);
1630 
1631 	btf = btf_get_from_fd(btf_fd, base_btf);
1632 	close(btf_fd);
1633 
1634 	return libbpf_ptr(btf);
1635 }
1636 
1637 struct btf *btf__load_from_kernel_by_id(__u32 id)
1638 {
1639 	return btf__load_from_kernel_by_id_split(id, NULL);
1640 }
1641 
1642 static void btf_invalidate_raw_data(struct btf *btf)
1643 {
1644 	if (btf->raw_data) {
1645 		free(btf->raw_data);
1646 		btf->raw_data = NULL;
1647 	}
1648 	if (btf->raw_data_swapped) {
1649 		free(btf->raw_data_swapped);
1650 		btf->raw_data_swapped = NULL;
1651 	}
1652 }
1653 
1654 /* Ensure BTF is ready to be modified (by splitting into a three memory
1655  * regions for header, types, and strings). Also invalidate cached
1656  * raw_data, if any.
1657  */
1658 static int btf_ensure_modifiable(struct btf *btf)
1659 {
1660 	void *hdr, *types;
1661 	struct strset *set = NULL;
1662 	int err = -ENOMEM;
1663 
1664 	if (btf_is_modifiable(btf)) {
1665 		/* any BTF modification invalidates raw_data */
1666 		btf_invalidate_raw_data(btf);
1667 		return 0;
1668 	}
1669 
1670 	/* split raw data into three memory regions */
1671 	hdr = malloc(btf->hdr->hdr_len);
1672 	types = malloc(btf->hdr->type_len);
1673 	if (!hdr || !types)
1674 		goto err_out;
1675 
1676 	memcpy(hdr, btf->hdr, btf->hdr->hdr_len);
1677 	memcpy(types, btf->types_data, btf->hdr->type_len);
1678 
1679 	/* build lookup index for all strings */
1680 	set = strset__new(BTF_MAX_STR_OFFSET, btf->strs_data, btf->hdr->str_len);
1681 	if (IS_ERR(set)) {
1682 		err = PTR_ERR(set);
1683 		goto err_out;
1684 	}
1685 
1686 	/* only when everything was successful, update internal state */
1687 	btf->hdr = hdr;
1688 	btf->types_data = types;
1689 	btf->types_data_cap = btf->hdr->type_len;
1690 	btf->strs_data = NULL;
1691 	btf->strs_set = set;
1692 	/* if BTF was created from scratch, all strings are guaranteed to be
1693 	 * unique and deduplicated
1694 	 */
1695 	if (btf->hdr->str_len == 0)
1696 		btf->strs_deduped = true;
1697 	if (!btf->base_btf && btf->hdr->str_len == 1)
1698 		btf->strs_deduped = true;
1699 
1700 	/* invalidate raw_data representation */
1701 	btf_invalidate_raw_data(btf);
1702 
1703 	return 0;
1704 
1705 err_out:
1706 	strset__free(set);
1707 	free(hdr);
1708 	free(types);
1709 	return err;
1710 }
1711 
1712 /* Find an offset in BTF string section that corresponds to a given string *s*.
1713  * Returns:
1714  *   - >0 offset into string section, if string is found;
1715  *   - -ENOENT, if string is not in the string section;
1716  *   - <0, on any other error.
1717  */
1718 int btf__find_str(struct btf *btf, const char *s)
1719 {
1720 	int off;
1721 
1722 	if (btf->base_btf) {
1723 		off = btf__find_str(btf->base_btf, s);
1724 		if (off != -ENOENT)
1725 			return off;
1726 	}
1727 
1728 	/* BTF needs to be in a modifiable state to build string lookup index */
1729 	if (btf_ensure_modifiable(btf))
1730 		return libbpf_err(-ENOMEM);
1731 
1732 	off = strset__find_str(btf->strs_set, s);
1733 	if (off < 0)
1734 		return libbpf_err(off);
1735 
1736 	return btf->start_str_off + off;
1737 }
1738 
1739 /* Add a string s to the BTF string section.
1740  * Returns:
1741  *   - > 0 offset into string section, on success;
1742  *   - < 0, on error.
1743  */
1744 int btf__add_str(struct btf *btf, const char *s)
1745 {
1746 	int off;
1747 
1748 	if (btf->base_btf) {
1749 		off = btf__find_str(btf->base_btf, s);
1750 		if (off != -ENOENT)
1751 			return off;
1752 	}
1753 
1754 	if (btf_ensure_modifiable(btf))
1755 		return libbpf_err(-ENOMEM);
1756 
1757 	off = strset__add_str(btf->strs_set, s);
1758 	if (off < 0)
1759 		return libbpf_err(off);
1760 
1761 	btf->hdr->str_len = strset__data_size(btf->strs_set);
1762 
1763 	return btf->start_str_off + off;
1764 }
1765 
1766 static void *btf_add_type_mem(struct btf *btf, size_t add_sz)
1767 {
1768 	return libbpf_add_mem(&btf->types_data, &btf->types_data_cap, 1,
1769 			      btf->hdr->type_len, UINT_MAX, add_sz);
1770 }
1771 
1772 static void btf_type_inc_vlen(struct btf_type *t)
1773 {
1774 	t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, btf_kflag(t));
1775 }
1776 
1777 static int btf_commit_type(struct btf *btf, int data_sz)
1778 {
1779 	int err;
1780 
1781 	err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1782 	if (err)
1783 		return libbpf_err(err);
1784 
1785 	btf->hdr->type_len += data_sz;
1786 	btf->hdr->str_off += data_sz;
1787 	btf->nr_types++;
1788 	return btf->start_id + btf->nr_types - 1;
1789 }
1790 
1791 struct btf_pipe {
1792 	const struct btf *src;
1793 	struct btf *dst;
1794 	struct hashmap *str_off_map; /* map string offsets from src to dst */
1795 };
1796 
1797 static int btf_rewrite_str(struct btf_pipe *p, __u32 *str_off)
1798 {
1799 	long mapped_off;
1800 	int off, err;
1801 
1802 	if (!*str_off) /* nothing to do for empty strings */
1803 		return 0;
1804 
1805 	if (p->str_off_map &&
1806 	    hashmap__find(p->str_off_map, *str_off, &mapped_off)) {
1807 		*str_off = mapped_off;
1808 		return 0;
1809 	}
1810 
1811 	off = btf__add_str(p->dst, btf__str_by_offset(p->src, *str_off));
1812 	if (off < 0)
1813 		return off;
1814 
1815 	/* Remember string mapping from src to dst.  It avoids
1816 	 * performing expensive string comparisons.
1817 	 */
1818 	if (p->str_off_map) {
1819 		err = hashmap__append(p->str_off_map, *str_off, off);
1820 		if (err)
1821 			return err;
1822 	}
1823 
1824 	*str_off = off;
1825 	return 0;
1826 }
1827 
1828 static int btf_add_type(struct btf_pipe *p, const struct btf_type *src_type)
1829 {
1830 	struct btf_field_iter it;
1831 	struct btf_type *t;
1832 	__u32 *str_off;
1833 	int sz, err;
1834 
1835 	sz = btf_type_size(src_type);
1836 	if (sz < 0)
1837 		return libbpf_err(sz);
1838 
1839 	/* deconstruct BTF, if necessary, and invalidate raw_data */
1840 	if (btf_ensure_modifiable(p->dst))
1841 		return libbpf_err(-ENOMEM);
1842 
1843 	t = btf_add_type_mem(p->dst, sz);
1844 	if (!t)
1845 		return libbpf_err(-ENOMEM);
1846 
1847 	memcpy(t, src_type, sz);
1848 
1849 	err = btf_field_iter_init(&it, t, BTF_FIELD_ITER_STRS);
1850 	if (err)
1851 		return libbpf_err(err);
1852 
1853 	while ((str_off = btf_field_iter_next(&it))) {
1854 		err = btf_rewrite_str(p, str_off);
1855 		if (err)
1856 			return libbpf_err(err);
1857 	}
1858 
1859 	return btf_commit_type(p->dst, sz);
1860 }
1861 
1862 int btf__add_type(struct btf *btf, const struct btf *src_btf, const struct btf_type *src_type)
1863 {
1864 	struct btf_pipe p = { .src = src_btf, .dst = btf };
1865 
1866 	return btf_add_type(&p, src_type);
1867 }
1868 
1869 static size_t btf_dedup_identity_hash_fn(long key, void *ctx);
1870 static bool btf_dedup_equal_fn(long k1, long k2, void *ctx);
1871 
1872 int btf__add_btf(struct btf *btf, const struct btf *src_btf)
1873 {
1874 	struct btf_pipe p = { .src = src_btf, .dst = btf };
1875 	int data_sz, sz, cnt, i, err, old_strs_len;
1876 	__u32 *off;
1877 	void *t;
1878 
1879 	/* appending split BTF isn't supported yet */
1880 	if (src_btf->base_btf)
1881 		return libbpf_err(-ENOTSUP);
1882 
1883 	/* deconstruct BTF, if necessary, and invalidate raw_data */
1884 	if (btf_ensure_modifiable(btf))
1885 		return libbpf_err(-ENOMEM);
1886 
1887 	/* remember original strings section size if we have to roll back
1888 	 * partial strings section changes
1889 	 */
1890 	old_strs_len = btf->hdr->str_len;
1891 
1892 	data_sz = src_btf->hdr->type_len;
1893 	cnt = btf__type_cnt(src_btf) - 1;
1894 
1895 	/* pre-allocate enough memory for new types */
1896 	t = btf_add_type_mem(btf, data_sz);
1897 	if (!t)
1898 		return libbpf_err(-ENOMEM);
1899 
1900 	/* pre-allocate enough memory for type offset index for new types */
1901 	off = btf_add_type_offs_mem(btf, cnt);
1902 	if (!off)
1903 		return libbpf_err(-ENOMEM);
1904 
1905 	/* Map the string offsets from src_btf to the offsets from btf to improve performance */
1906 	p.str_off_map = hashmap__new(btf_dedup_identity_hash_fn, btf_dedup_equal_fn, NULL);
1907 	if (IS_ERR(p.str_off_map))
1908 		return libbpf_err(-ENOMEM);
1909 
1910 	/* bulk copy types data for all types from src_btf */
1911 	memcpy(t, src_btf->types_data, data_sz);
1912 
1913 	for (i = 0; i < cnt; i++) {
1914 		struct btf_field_iter it;
1915 		__u32 *type_id, *str_off;
1916 
1917 		sz = btf_type_size(t);
1918 		if (sz < 0) {
1919 			/* unlikely, has to be corrupted src_btf */
1920 			err = sz;
1921 			goto err_out;
1922 		}
1923 
1924 		/* fill out type ID to type offset mapping for lookups by type ID */
1925 		*off = t - btf->types_data;
1926 
1927 		/* add, dedup, and remap strings referenced by this BTF type */
1928 		err = btf_field_iter_init(&it, t, BTF_FIELD_ITER_STRS);
1929 		if (err)
1930 			goto err_out;
1931 		while ((str_off = btf_field_iter_next(&it))) {
1932 			err = btf_rewrite_str(&p, str_off);
1933 			if (err)
1934 				goto err_out;
1935 		}
1936 
1937 		/* remap all type IDs referenced from this BTF type */
1938 		err = btf_field_iter_init(&it, t, BTF_FIELD_ITER_IDS);
1939 		if (err)
1940 			goto err_out;
1941 
1942 		while ((type_id = btf_field_iter_next(&it))) {
1943 			if (!*type_id) /* nothing to do for VOID references */
1944 				continue;
1945 
1946 			/* we haven't updated btf's type count yet, so
1947 			 * btf->start_id + btf->nr_types - 1 is the type ID offset we should
1948 			 * add to all newly added BTF types
1949 			 */
1950 			*type_id += btf->start_id + btf->nr_types - 1;
1951 		}
1952 
1953 		/* go to next type data and type offset index entry */
1954 		t += sz;
1955 		off++;
1956 	}
1957 
1958 	/* Up until now any of the copied type data was effectively invisible,
1959 	 * so if we exited early before this point due to error, BTF would be
1960 	 * effectively unmodified. There would be extra internal memory
1961 	 * pre-allocated, but it would not be available for querying.  But now
1962 	 * that we've copied and rewritten all the data successfully, we can
1963 	 * update type count and various internal offsets and sizes to
1964 	 * "commit" the changes and made them visible to the outside world.
1965 	 */
1966 	btf->hdr->type_len += data_sz;
1967 	btf->hdr->str_off += data_sz;
1968 	btf->nr_types += cnt;
1969 
1970 	hashmap__free(p.str_off_map);
1971 
1972 	/* return type ID of the first added BTF type */
1973 	return btf->start_id + btf->nr_types - cnt;
1974 err_out:
1975 	/* zero out preallocated memory as if it was just allocated with
1976 	 * libbpf_add_mem()
1977 	 */
1978 	memset(btf->types_data + btf->hdr->type_len, 0, data_sz);
1979 	memset(btf->strs_data + old_strs_len, 0, btf->hdr->str_len - old_strs_len);
1980 
1981 	/* and now restore original strings section size; types data size
1982 	 * wasn't modified, so doesn't need restoring, see big comment above
1983 	 */
1984 	btf->hdr->str_len = old_strs_len;
1985 
1986 	hashmap__free(p.str_off_map);
1987 
1988 	return libbpf_err(err);
1989 }
1990 
1991 /*
1992  * Append new BTF_KIND_INT type with:
1993  *   - *name* - non-empty, non-NULL type name;
1994  *   - *sz* - power-of-2 (1, 2, 4, ..) size of the type, in bytes;
1995  *   - encoding is a combination of BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL.
1996  * Returns:
1997  *   - >0, type ID of newly added BTF type;
1998  *   - <0, on error.
1999  */
2000 int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding)
2001 {
2002 	struct btf_type *t;
2003 	int sz, name_off;
2004 
2005 	/* non-empty name */
2006 	if (!name || !name[0])
2007 		return libbpf_err(-EINVAL);
2008 	/* byte_sz must be power of 2 */
2009 	if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
2010 		return libbpf_err(-EINVAL);
2011 	if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
2012 		return libbpf_err(-EINVAL);
2013 
2014 	/* deconstruct BTF, if necessary, and invalidate raw_data */
2015 	if (btf_ensure_modifiable(btf))
2016 		return libbpf_err(-ENOMEM);
2017 
2018 	sz = sizeof(struct btf_type) + sizeof(int);
2019 	t = btf_add_type_mem(btf, sz);
2020 	if (!t)
2021 		return libbpf_err(-ENOMEM);
2022 
2023 	/* if something goes wrong later, we might end up with an extra string,
2024 	 * but that shouldn't be a problem, because BTF can't be constructed
2025 	 * completely anyway and will most probably be just discarded
2026 	 */
2027 	name_off = btf__add_str(btf, name);
2028 	if (name_off < 0)
2029 		return name_off;
2030 
2031 	t->name_off = name_off;
2032 	t->info = btf_type_info(BTF_KIND_INT, 0, 0);
2033 	t->size = byte_sz;
2034 	/* set INT info, we don't allow setting legacy bit offset/size */
2035 	*(__u32 *)(t + 1) = (encoding << 24) | (byte_sz * 8);
2036 
2037 	return btf_commit_type(btf, sz);
2038 }
2039 
2040 /*
2041  * Append new BTF_KIND_FLOAT type with:
2042  *   - *name* - non-empty, non-NULL type name;
2043  *   - *sz* - size of the type, in bytes;
2044  * Returns:
2045  *   - >0, type ID of newly added BTF type;
2046  *   - <0, on error.
2047  */
2048 int btf__add_float(struct btf *btf, const char *name, size_t byte_sz)
2049 {
2050 	struct btf_type *t;
2051 	int sz, name_off;
2052 
2053 	/* non-empty name */
2054 	if (!name || !name[0])
2055 		return libbpf_err(-EINVAL);
2056 
2057 	/* byte_sz must be one of the explicitly allowed values */
2058 	if (byte_sz != 2 && byte_sz != 4 && byte_sz != 8 && byte_sz != 12 &&
2059 	    byte_sz != 16)
2060 		return libbpf_err(-EINVAL);
2061 
2062 	if (btf_ensure_modifiable(btf))
2063 		return libbpf_err(-ENOMEM);
2064 
2065 	sz = sizeof(struct btf_type);
2066 	t = btf_add_type_mem(btf, sz);
2067 	if (!t)
2068 		return libbpf_err(-ENOMEM);
2069 
2070 	name_off = btf__add_str(btf, name);
2071 	if (name_off < 0)
2072 		return name_off;
2073 
2074 	t->name_off = name_off;
2075 	t->info = btf_type_info(BTF_KIND_FLOAT, 0, 0);
2076 	t->size = byte_sz;
2077 
2078 	return btf_commit_type(btf, sz);
2079 }
2080 
2081 /* it's completely legal to append BTF types with type IDs pointing forward to
2082  * types that haven't been appended yet, so we only make sure that id looks
2083  * sane, we can't guarantee that ID will always be valid
2084  */
2085 static int validate_type_id(int id)
2086 {
2087 	if (id < 0 || id > BTF_MAX_NR_TYPES)
2088 		return -EINVAL;
2089 	return 0;
2090 }
2091 
2092 /* generic append function for PTR, TYPEDEF, CONST/VOLATILE/RESTRICT */
2093 static int btf_add_ref_kind(struct btf *btf, int kind, const char *name, int ref_type_id, int kflag)
2094 {
2095 	struct btf_type *t;
2096 	int sz, name_off = 0;
2097 
2098 	if (validate_type_id(ref_type_id))
2099 		return libbpf_err(-EINVAL);
2100 
2101 	if (btf_ensure_modifiable(btf))
2102 		return libbpf_err(-ENOMEM);
2103 
2104 	sz = sizeof(struct btf_type);
2105 	t = btf_add_type_mem(btf, sz);
2106 	if (!t)
2107 		return libbpf_err(-ENOMEM);
2108 
2109 	if (name && name[0]) {
2110 		name_off = btf__add_str(btf, name);
2111 		if (name_off < 0)
2112 			return name_off;
2113 	}
2114 
2115 	t->name_off = name_off;
2116 	t->info = btf_type_info(kind, 0, kflag);
2117 	t->type = ref_type_id;
2118 
2119 	return btf_commit_type(btf, sz);
2120 }
2121 
2122 /*
2123  * Append new BTF_KIND_PTR type with:
2124  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2125  * Returns:
2126  *   - >0, type ID of newly added BTF type;
2127  *   - <0, on error.
2128  */
2129 int btf__add_ptr(struct btf *btf, int ref_type_id)
2130 {
2131 	return btf_add_ref_kind(btf, BTF_KIND_PTR, NULL, ref_type_id, 0);
2132 }
2133 
2134 /*
2135  * Append new BTF_KIND_ARRAY type with:
2136  *   - *index_type_id* - type ID of the type describing array index;
2137  *   - *elem_type_id* - type ID of the type describing array element;
2138  *   - *nr_elems* - the size of the array;
2139  * Returns:
2140  *   - >0, type ID of newly added BTF type;
2141  *   - <0, on error.
2142  */
2143 int btf__add_array(struct btf *btf, int index_type_id, int elem_type_id, __u32 nr_elems)
2144 {
2145 	struct btf_type *t;
2146 	struct btf_array *a;
2147 	int sz;
2148 
2149 	if (validate_type_id(index_type_id) || validate_type_id(elem_type_id))
2150 		return libbpf_err(-EINVAL);
2151 
2152 	if (btf_ensure_modifiable(btf))
2153 		return libbpf_err(-ENOMEM);
2154 
2155 	sz = sizeof(struct btf_type) + sizeof(struct btf_array);
2156 	t = btf_add_type_mem(btf, sz);
2157 	if (!t)
2158 		return libbpf_err(-ENOMEM);
2159 
2160 	t->name_off = 0;
2161 	t->info = btf_type_info(BTF_KIND_ARRAY, 0, 0);
2162 	t->size = 0;
2163 
2164 	a = btf_array(t);
2165 	a->type = elem_type_id;
2166 	a->index_type = index_type_id;
2167 	a->nelems = nr_elems;
2168 
2169 	return btf_commit_type(btf, sz);
2170 }
2171 
2172 /* generic STRUCT/UNION append function */
2173 static int btf_add_composite(struct btf *btf, int kind, const char *name, __u32 bytes_sz)
2174 {
2175 	struct btf_type *t;
2176 	int sz, name_off = 0;
2177 
2178 	if (btf_ensure_modifiable(btf))
2179 		return libbpf_err(-ENOMEM);
2180 
2181 	sz = sizeof(struct btf_type);
2182 	t = btf_add_type_mem(btf, sz);
2183 	if (!t)
2184 		return libbpf_err(-ENOMEM);
2185 
2186 	if (name && name[0]) {
2187 		name_off = btf__add_str(btf, name);
2188 		if (name_off < 0)
2189 			return name_off;
2190 	}
2191 
2192 	/* start out with vlen=0 and no kflag; this will be adjusted when
2193 	 * adding each member
2194 	 */
2195 	t->name_off = name_off;
2196 	t->info = btf_type_info(kind, 0, 0);
2197 	t->size = bytes_sz;
2198 
2199 	return btf_commit_type(btf, sz);
2200 }
2201 
2202 /*
2203  * Append new BTF_KIND_STRUCT type with:
2204  *   - *name* - name of the struct, can be NULL or empty for anonymous structs;
2205  *   - *byte_sz* - size of the struct, in bytes;
2206  *
2207  * Struct initially has no fields in it. Fields can be added by
2208  * btf__add_field() right after btf__add_struct() succeeds.
2209  *
2210  * Returns:
2211  *   - >0, type ID of newly added BTF type;
2212  *   - <0, on error.
2213  */
2214 int btf__add_struct(struct btf *btf, const char *name, __u32 byte_sz)
2215 {
2216 	return btf_add_composite(btf, BTF_KIND_STRUCT, name, byte_sz);
2217 }
2218 
2219 /*
2220  * Append new BTF_KIND_UNION type with:
2221  *   - *name* - name of the union, can be NULL or empty for anonymous union;
2222  *   - *byte_sz* - size of the union, in bytes;
2223  *
2224  * Union initially has no fields in it. Fields can be added by
2225  * btf__add_field() right after btf__add_union() succeeds. All fields
2226  * should have *bit_offset* of 0.
2227  *
2228  * Returns:
2229  *   - >0, type ID of newly added BTF type;
2230  *   - <0, on error.
2231  */
2232 int btf__add_union(struct btf *btf, const char *name, __u32 byte_sz)
2233 {
2234 	return btf_add_composite(btf, BTF_KIND_UNION, name, byte_sz);
2235 }
2236 
2237 static struct btf_type *btf_last_type(struct btf *btf)
2238 {
2239 	return btf_type_by_id(btf, btf__type_cnt(btf) - 1);
2240 }
2241 
2242 /*
2243  * Append new field for the current STRUCT/UNION type with:
2244  *   - *name* - name of the field, can be NULL or empty for anonymous field;
2245  *   - *type_id* - type ID for the type describing field type;
2246  *   - *bit_offset* - bit offset of the start of the field within struct/union;
2247  *   - *bit_size* - bit size of a bitfield, 0 for non-bitfield fields;
2248  * Returns:
2249  *   -  0, on success;
2250  *   - <0, on error.
2251  */
2252 int btf__add_field(struct btf *btf, const char *name, int type_id,
2253 		   __u32 bit_offset, __u32 bit_size)
2254 {
2255 	struct btf_type *t;
2256 	struct btf_member *m;
2257 	bool is_bitfield;
2258 	int sz, name_off = 0;
2259 
2260 	/* last type should be union/struct */
2261 	if (btf->nr_types == 0)
2262 		return libbpf_err(-EINVAL);
2263 	t = btf_last_type(btf);
2264 	if (!btf_is_composite(t))
2265 		return libbpf_err(-EINVAL);
2266 
2267 	if (validate_type_id(type_id))
2268 		return libbpf_err(-EINVAL);
2269 	/* best-effort bit field offset/size enforcement */
2270 	is_bitfield = bit_size || (bit_offset % 8 != 0);
2271 	if (is_bitfield && (bit_size == 0 || bit_size > 255 || bit_offset > 0xffffff))
2272 		return libbpf_err(-EINVAL);
2273 
2274 	/* only offset 0 is allowed for unions */
2275 	if (btf_is_union(t) && bit_offset)
2276 		return libbpf_err(-EINVAL);
2277 
2278 	/* decompose and invalidate raw data */
2279 	if (btf_ensure_modifiable(btf))
2280 		return libbpf_err(-ENOMEM);
2281 
2282 	sz = sizeof(struct btf_member);
2283 	m = btf_add_type_mem(btf, sz);
2284 	if (!m)
2285 		return libbpf_err(-ENOMEM);
2286 
2287 	if (name && name[0]) {
2288 		name_off = btf__add_str(btf, name);
2289 		if (name_off < 0)
2290 			return name_off;
2291 	}
2292 
2293 	m->name_off = name_off;
2294 	m->type = type_id;
2295 	m->offset = bit_offset | (bit_size << 24);
2296 
2297 	/* btf_add_type_mem can invalidate t pointer */
2298 	t = btf_last_type(btf);
2299 	/* update parent type's vlen and kflag */
2300 	t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, is_bitfield || btf_kflag(t));
2301 
2302 	btf->hdr->type_len += sz;
2303 	btf->hdr->str_off += sz;
2304 	return 0;
2305 }
2306 
2307 static int btf_add_enum_common(struct btf *btf, const char *name, __u32 byte_sz,
2308 			       bool is_signed, __u8 kind)
2309 {
2310 	struct btf_type *t;
2311 	int sz, name_off = 0;
2312 
2313 	/* byte_sz must be power of 2 */
2314 	if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 8)
2315 		return libbpf_err(-EINVAL);
2316 
2317 	if (btf_ensure_modifiable(btf))
2318 		return libbpf_err(-ENOMEM);
2319 
2320 	sz = sizeof(struct btf_type);
2321 	t = btf_add_type_mem(btf, sz);
2322 	if (!t)
2323 		return libbpf_err(-ENOMEM);
2324 
2325 	if (name && name[0]) {
2326 		name_off = btf__add_str(btf, name);
2327 		if (name_off < 0)
2328 			return name_off;
2329 	}
2330 
2331 	/* start out with vlen=0; it will be adjusted when adding enum values */
2332 	t->name_off = name_off;
2333 	t->info = btf_type_info(kind, 0, is_signed);
2334 	t->size = byte_sz;
2335 
2336 	return btf_commit_type(btf, sz);
2337 }
2338 
2339 /*
2340  * Append new BTF_KIND_ENUM type with:
2341  *   - *name* - name of the enum, can be NULL or empty for anonymous enums;
2342  *   - *byte_sz* - size of the enum, in bytes.
2343  *
2344  * Enum initially has no enum values in it (and corresponds to enum forward
2345  * declaration). Enumerator values can be added by btf__add_enum_value()
2346  * immediately after btf__add_enum() succeeds.
2347  *
2348  * Returns:
2349  *   - >0, type ID of newly added BTF type;
2350  *   - <0, on error.
2351  */
2352 int btf__add_enum(struct btf *btf, const char *name, __u32 byte_sz)
2353 {
2354 	/*
2355 	 * set the signedness to be unsigned, it will change to signed
2356 	 * if any later enumerator is negative.
2357 	 */
2358 	return btf_add_enum_common(btf, name, byte_sz, false, BTF_KIND_ENUM);
2359 }
2360 
2361 /*
2362  * Append new enum value for the current ENUM type with:
2363  *   - *name* - name of the enumerator value, can't be NULL or empty;
2364  *   - *value* - integer value corresponding to enum value *name*;
2365  * Returns:
2366  *   -  0, on success;
2367  *   - <0, on error.
2368  */
2369 int btf__add_enum_value(struct btf *btf, const char *name, __s64 value)
2370 {
2371 	struct btf_type *t;
2372 	struct btf_enum *v;
2373 	int sz, name_off;
2374 
2375 	/* last type should be BTF_KIND_ENUM */
2376 	if (btf->nr_types == 0)
2377 		return libbpf_err(-EINVAL);
2378 	t = btf_last_type(btf);
2379 	if (!btf_is_enum(t))
2380 		return libbpf_err(-EINVAL);
2381 
2382 	/* non-empty name */
2383 	if (!name || !name[0])
2384 		return libbpf_err(-EINVAL);
2385 	if (value < INT_MIN || value > UINT_MAX)
2386 		return libbpf_err(-E2BIG);
2387 
2388 	/* decompose and invalidate raw data */
2389 	if (btf_ensure_modifiable(btf))
2390 		return libbpf_err(-ENOMEM);
2391 
2392 	sz = sizeof(struct btf_enum);
2393 	v = btf_add_type_mem(btf, sz);
2394 	if (!v)
2395 		return libbpf_err(-ENOMEM);
2396 
2397 	name_off = btf__add_str(btf, name);
2398 	if (name_off < 0)
2399 		return name_off;
2400 
2401 	v->name_off = name_off;
2402 	v->val = value;
2403 
2404 	/* update parent type's vlen */
2405 	t = btf_last_type(btf);
2406 	btf_type_inc_vlen(t);
2407 
2408 	/* if negative value, set signedness to signed */
2409 	if (value < 0)
2410 		t->info = btf_type_info(btf_kind(t), btf_vlen(t), true);
2411 
2412 	btf->hdr->type_len += sz;
2413 	btf->hdr->str_off += sz;
2414 	return 0;
2415 }
2416 
2417 /*
2418  * Append new BTF_KIND_ENUM64 type with:
2419  *   - *name* - name of the enum, can be NULL or empty for anonymous enums;
2420  *   - *byte_sz* - size of the enum, in bytes.
2421  *   - *is_signed* - whether the enum values are signed or not;
2422  *
2423  * Enum initially has no enum values in it (and corresponds to enum forward
2424  * declaration). Enumerator values can be added by btf__add_enum64_value()
2425  * immediately after btf__add_enum64() succeeds.
2426  *
2427  * Returns:
2428  *   - >0, type ID of newly added BTF type;
2429  *   - <0, on error.
2430  */
2431 int btf__add_enum64(struct btf *btf, const char *name, __u32 byte_sz,
2432 		    bool is_signed)
2433 {
2434 	return btf_add_enum_common(btf, name, byte_sz, is_signed,
2435 				   BTF_KIND_ENUM64);
2436 }
2437 
2438 /*
2439  * Append new enum value for the current ENUM64 type with:
2440  *   - *name* - name of the enumerator value, can't be NULL or empty;
2441  *   - *value* - integer value corresponding to enum value *name*;
2442  * Returns:
2443  *   -  0, on success;
2444  *   - <0, on error.
2445  */
2446 int btf__add_enum64_value(struct btf *btf, const char *name, __u64 value)
2447 {
2448 	struct btf_enum64 *v;
2449 	struct btf_type *t;
2450 	int sz, name_off;
2451 
2452 	/* last type should be BTF_KIND_ENUM64 */
2453 	if (btf->nr_types == 0)
2454 		return libbpf_err(-EINVAL);
2455 	t = btf_last_type(btf);
2456 	if (!btf_is_enum64(t))
2457 		return libbpf_err(-EINVAL);
2458 
2459 	/* non-empty name */
2460 	if (!name || !name[0])
2461 		return libbpf_err(-EINVAL);
2462 
2463 	/* decompose and invalidate raw data */
2464 	if (btf_ensure_modifiable(btf))
2465 		return libbpf_err(-ENOMEM);
2466 
2467 	sz = sizeof(struct btf_enum64);
2468 	v = btf_add_type_mem(btf, sz);
2469 	if (!v)
2470 		return libbpf_err(-ENOMEM);
2471 
2472 	name_off = btf__add_str(btf, name);
2473 	if (name_off < 0)
2474 		return name_off;
2475 
2476 	v->name_off = name_off;
2477 	v->val_lo32 = (__u32)value;
2478 	v->val_hi32 = value >> 32;
2479 
2480 	/* update parent type's vlen */
2481 	t = btf_last_type(btf);
2482 	btf_type_inc_vlen(t);
2483 
2484 	btf->hdr->type_len += sz;
2485 	btf->hdr->str_off += sz;
2486 	return 0;
2487 }
2488 
2489 /*
2490  * Append new BTF_KIND_FWD type with:
2491  *   - *name*, non-empty/non-NULL name;
2492  *   - *fwd_kind*, kind of forward declaration, one of BTF_FWD_STRUCT,
2493  *     BTF_FWD_UNION, or BTF_FWD_ENUM;
2494  * Returns:
2495  *   - >0, type ID of newly added BTF type;
2496  *   - <0, on error.
2497  */
2498 int btf__add_fwd(struct btf *btf, const char *name, enum btf_fwd_kind fwd_kind)
2499 {
2500 	if (!name || !name[0])
2501 		return libbpf_err(-EINVAL);
2502 
2503 	switch (fwd_kind) {
2504 	case BTF_FWD_STRUCT:
2505 	case BTF_FWD_UNION: {
2506 		struct btf_type *t;
2507 		int id;
2508 
2509 		id = btf_add_ref_kind(btf, BTF_KIND_FWD, name, 0, 0);
2510 		if (id <= 0)
2511 			return id;
2512 		t = btf_type_by_id(btf, id);
2513 		t->info = btf_type_info(BTF_KIND_FWD, 0, fwd_kind == BTF_FWD_UNION);
2514 		return id;
2515 	}
2516 	case BTF_FWD_ENUM:
2517 		/* enum forward in BTF currently is just an enum with no enum
2518 		 * values; we also assume a standard 4-byte size for it
2519 		 */
2520 		return btf__add_enum(btf, name, sizeof(int));
2521 	default:
2522 		return libbpf_err(-EINVAL);
2523 	}
2524 }
2525 
2526 /*
2527  * Append new BTF_KING_TYPEDEF type with:
2528  *   - *name*, non-empty/non-NULL name;
2529  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2530  * Returns:
2531  *   - >0, type ID of newly added BTF type;
2532  *   - <0, on error.
2533  */
2534 int btf__add_typedef(struct btf *btf, const char *name, int ref_type_id)
2535 {
2536 	if (!name || !name[0])
2537 		return libbpf_err(-EINVAL);
2538 
2539 	return btf_add_ref_kind(btf, BTF_KIND_TYPEDEF, name, ref_type_id, 0);
2540 }
2541 
2542 /*
2543  * Append new BTF_KIND_VOLATILE type with:
2544  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2545  * Returns:
2546  *   - >0, type ID of newly added BTF type;
2547  *   - <0, on error.
2548  */
2549 int btf__add_volatile(struct btf *btf, int ref_type_id)
2550 {
2551 	return btf_add_ref_kind(btf, BTF_KIND_VOLATILE, NULL, ref_type_id, 0);
2552 }
2553 
2554 /*
2555  * Append new BTF_KIND_CONST type with:
2556  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2557  * Returns:
2558  *   - >0, type ID of newly added BTF type;
2559  *   - <0, on error.
2560  */
2561 int btf__add_const(struct btf *btf, int ref_type_id)
2562 {
2563 	return btf_add_ref_kind(btf, BTF_KIND_CONST, NULL, ref_type_id, 0);
2564 }
2565 
2566 /*
2567  * Append new BTF_KIND_RESTRICT type with:
2568  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2569  * Returns:
2570  *   - >0, type ID of newly added BTF type;
2571  *   - <0, on error.
2572  */
2573 int btf__add_restrict(struct btf *btf, int ref_type_id)
2574 {
2575 	return btf_add_ref_kind(btf, BTF_KIND_RESTRICT, NULL, ref_type_id, 0);
2576 }
2577 
2578 /*
2579  * Append new BTF_KIND_TYPE_TAG type with:
2580  *   - *value*, non-empty/non-NULL tag value;
2581  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2582  * Returns:
2583  *   - >0, type ID of newly added BTF type;
2584  *   - <0, on error.
2585  */
2586 int btf__add_type_tag(struct btf *btf, const char *value, int ref_type_id)
2587 {
2588 	if (!value || !value[0])
2589 		return libbpf_err(-EINVAL);
2590 
2591 	return btf_add_ref_kind(btf, BTF_KIND_TYPE_TAG, value, ref_type_id, 0);
2592 }
2593 
2594 /*
2595  * Append new BTF_KIND_TYPE_TAG type with:
2596  *   - *value*, non-empty/non-NULL tag value;
2597  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2598  * Set info->kflag to 1, indicating this tag is an __attribute__
2599  * Returns:
2600  *   - >0, type ID of newly added BTF type;
2601  *   - <0, on error.
2602  */
2603 int btf__add_type_attr(struct btf *btf, const char *value, int ref_type_id)
2604 {
2605 	if (!value || !value[0])
2606 		return libbpf_err(-EINVAL);
2607 
2608 	return btf_add_ref_kind(btf, BTF_KIND_TYPE_TAG, value, ref_type_id, 1);
2609 }
2610 
2611 /*
2612  * Append new BTF_KIND_FUNC type with:
2613  *   - *name*, non-empty/non-NULL name;
2614  *   - *proto_type_id* - FUNC_PROTO's type ID, it might not exist yet;
2615  * Returns:
2616  *   - >0, type ID of newly added BTF type;
2617  *   - <0, on error.
2618  */
2619 int btf__add_func(struct btf *btf, const char *name,
2620 		  enum btf_func_linkage linkage, int proto_type_id)
2621 {
2622 	int id;
2623 
2624 	if (!name || !name[0])
2625 		return libbpf_err(-EINVAL);
2626 	if (linkage != BTF_FUNC_STATIC && linkage != BTF_FUNC_GLOBAL &&
2627 	    linkage != BTF_FUNC_EXTERN)
2628 		return libbpf_err(-EINVAL);
2629 
2630 	id = btf_add_ref_kind(btf, BTF_KIND_FUNC, name, proto_type_id, 0);
2631 	if (id > 0) {
2632 		struct btf_type *t = btf_type_by_id(btf, id);
2633 
2634 		t->info = btf_type_info(BTF_KIND_FUNC, linkage, 0);
2635 	}
2636 	return libbpf_err(id);
2637 }
2638 
2639 /*
2640  * Append new BTF_KIND_FUNC_PROTO with:
2641  *   - *ret_type_id* - type ID for return result of a function.
2642  *
2643  * Function prototype initially has no arguments, but they can be added by
2644  * btf__add_func_param() one by one, immediately after
2645  * btf__add_func_proto() succeeded.
2646  *
2647  * Returns:
2648  *   - >0, type ID of newly added BTF type;
2649  *   - <0, on error.
2650  */
2651 int btf__add_func_proto(struct btf *btf, int ret_type_id)
2652 {
2653 	struct btf_type *t;
2654 	int sz;
2655 
2656 	if (validate_type_id(ret_type_id))
2657 		return libbpf_err(-EINVAL);
2658 
2659 	if (btf_ensure_modifiable(btf))
2660 		return libbpf_err(-ENOMEM);
2661 
2662 	sz = sizeof(struct btf_type);
2663 	t = btf_add_type_mem(btf, sz);
2664 	if (!t)
2665 		return libbpf_err(-ENOMEM);
2666 
2667 	/* start out with vlen=0; this will be adjusted when adding enum
2668 	 * values, if necessary
2669 	 */
2670 	t->name_off = 0;
2671 	t->info = btf_type_info(BTF_KIND_FUNC_PROTO, 0, 0);
2672 	t->type = ret_type_id;
2673 
2674 	return btf_commit_type(btf, sz);
2675 }
2676 
2677 /*
2678  * Append new function parameter for current FUNC_PROTO type with:
2679  *   - *name* - parameter name, can be NULL or empty;
2680  *   - *type_id* - type ID describing the type of the parameter.
2681  * Returns:
2682  *   -  0, on success;
2683  *   - <0, on error.
2684  */
2685 int btf__add_func_param(struct btf *btf, const char *name, int type_id)
2686 {
2687 	struct btf_type *t;
2688 	struct btf_param *p;
2689 	int sz, name_off = 0;
2690 
2691 	if (validate_type_id(type_id))
2692 		return libbpf_err(-EINVAL);
2693 
2694 	/* last type should be BTF_KIND_FUNC_PROTO */
2695 	if (btf->nr_types == 0)
2696 		return libbpf_err(-EINVAL);
2697 	t = btf_last_type(btf);
2698 	if (!btf_is_func_proto(t))
2699 		return libbpf_err(-EINVAL);
2700 
2701 	/* decompose and invalidate raw data */
2702 	if (btf_ensure_modifiable(btf))
2703 		return libbpf_err(-ENOMEM);
2704 
2705 	sz = sizeof(struct btf_param);
2706 	p = btf_add_type_mem(btf, sz);
2707 	if (!p)
2708 		return libbpf_err(-ENOMEM);
2709 
2710 	if (name && name[0]) {
2711 		name_off = btf__add_str(btf, name);
2712 		if (name_off < 0)
2713 			return name_off;
2714 	}
2715 
2716 	p->name_off = name_off;
2717 	p->type = type_id;
2718 
2719 	/* update parent type's vlen */
2720 	t = btf_last_type(btf);
2721 	btf_type_inc_vlen(t);
2722 
2723 	btf->hdr->type_len += sz;
2724 	btf->hdr->str_off += sz;
2725 	return 0;
2726 }
2727 
2728 /*
2729  * Append new BTF_KIND_VAR type with:
2730  *   - *name* - non-empty/non-NULL name;
2731  *   - *linkage* - variable linkage, one of BTF_VAR_STATIC,
2732  *     BTF_VAR_GLOBAL_ALLOCATED, or BTF_VAR_GLOBAL_EXTERN;
2733  *   - *type_id* - type ID of the type describing the type of the variable.
2734  * Returns:
2735  *   - >0, type ID of newly added BTF type;
2736  *   - <0, on error.
2737  */
2738 int btf__add_var(struct btf *btf, const char *name, int linkage, int type_id)
2739 {
2740 	struct btf_type *t;
2741 	struct btf_var *v;
2742 	int sz, name_off;
2743 
2744 	/* non-empty name */
2745 	if (!name || !name[0])
2746 		return libbpf_err(-EINVAL);
2747 	if (linkage != BTF_VAR_STATIC && linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2748 	    linkage != BTF_VAR_GLOBAL_EXTERN)
2749 		return libbpf_err(-EINVAL);
2750 	if (validate_type_id(type_id))
2751 		return libbpf_err(-EINVAL);
2752 
2753 	/* deconstruct BTF, if necessary, and invalidate raw_data */
2754 	if (btf_ensure_modifiable(btf))
2755 		return libbpf_err(-ENOMEM);
2756 
2757 	sz = sizeof(struct btf_type) + sizeof(struct btf_var);
2758 	t = btf_add_type_mem(btf, sz);
2759 	if (!t)
2760 		return libbpf_err(-ENOMEM);
2761 
2762 	name_off = btf__add_str(btf, name);
2763 	if (name_off < 0)
2764 		return name_off;
2765 
2766 	t->name_off = name_off;
2767 	t->info = btf_type_info(BTF_KIND_VAR, 0, 0);
2768 	t->type = type_id;
2769 
2770 	v = btf_var(t);
2771 	v->linkage = linkage;
2772 
2773 	return btf_commit_type(btf, sz);
2774 }
2775 
2776 /*
2777  * Append new BTF_KIND_DATASEC type with:
2778  *   - *name* - non-empty/non-NULL name;
2779  *   - *byte_sz* - data section size, in bytes.
2780  *
2781  * Data section is initially empty. Variables info can be added with
2782  * btf__add_datasec_var_info() calls, after btf__add_datasec() succeeds.
2783  *
2784  * Returns:
2785  *   - >0, type ID of newly added BTF type;
2786  *   - <0, on error.
2787  */
2788 int btf__add_datasec(struct btf *btf, const char *name, __u32 byte_sz)
2789 {
2790 	struct btf_type *t;
2791 	int sz, name_off;
2792 
2793 	/* non-empty name */
2794 	if (!name || !name[0])
2795 		return libbpf_err(-EINVAL);
2796 
2797 	if (btf_ensure_modifiable(btf))
2798 		return libbpf_err(-ENOMEM);
2799 
2800 	sz = sizeof(struct btf_type);
2801 	t = btf_add_type_mem(btf, sz);
2802 	if (!t)
2803 		return libbpf_err(-ENOMEM);
2804 
2805 	name_off = btf__add_str(btf, name);
2806 	if (name_off < 0)
2807 		return name_off;
2808 
2809 	/* start with vlen=0, which will be update as var_secinfos are added */
2810 	t->name_off = name_off;
2811 	t->info = btf_type_info(BTF_KIND_DATASEC, 0, 0);
2812 	t->size = byte_sz;
2813 
2814 	return btf_commit_type(btf, sz);
2815 }
2816 
2817 /*
2818  * Append new data section variable information entry for current DATASEC type:
2819  *   - *var_type_id* - type ID, describing type of the variable;
2820  *   - *offset* - variable offset within data section, in bytes;
2821  *   - *byte_sz* - variable size, in bytes.
2822  *
2823  * Returns:
2824  *   -  0, on success;
2825  *   - <0, on error.
2826  */
2827 int btf__add_datasec_var_info(struct btf *btf, int var_type_id, __u32 offset, __u32 byte_sz)
2828 {
2829 	struct btf_type *t;
2830 	struct btf_var_secinfo *v;
2831 	int sz;
2832 
2833 	/* last type should be BTF_KIND_DATASEC */
2834 	if (btf->nr_types == 0)
2835 		return libbpf_err(-EINVAL);
2836 	t = btf_last_type(btf);
2837 	if (!btf_is_datasec(t))
2838 		return libbpf_err(-EINVAL);
2839 
2840 	if (validate_type_id(var_type_id))
2841 		return libbpf_err(-EINVAL);
2842 
2843 	/* decompose and invalidate raw data */
2844 	if (btf_ensure_modifiable(btf))
2845 		return libbpf_err(-ENOMEM);
2846 
2847 	sz = sizeof(struct btf_var_secinfo);
2848 	v = btf_add_type_mem(btf, sz);
2849 	if (!v)
2850 		return libbpf_err(-ENOMEM);
2851 
2852 	v->type = var_type_id;
2853 	v->offset = offset;
2854 	v->size = byte_sz;
2855 
2856 	/* update parent type's vlen */
2857 	t = btf_last_type(btf);
2858 	btf_type_inc_vlen(t);
2859 
2860 	btf->hdr->type_len += sz;
2861 	btf->hdr->str_off += sz;
2862 	return 0;
2863 }
2864 
2865 static int btf_add_decl_tag(struct btf *btf, const char *value, int ref_type_id,
2866 			    int component_idx, int kflag)
2867 {
2868 	struct btf_type *t;
2869 	int sz, value_off;
2870 
2871 	if (!value || !value[0] || component_idx < -1)
2872 		return libbpf_err(-EINVAL);
2873 
2874 	if (validate_type_id(ref_type_id))
2875 		return libbpf_err(-EINVAL);
2876 
2877 	if (btf_ensure_modifiable(btf))
2878 		return libbpf_err(-ENOMEM);
2879 
2880 	sz = sizeof(struct btf_type) + sizeof(struct btf_decl_tag);
2881 	t = btf_add_type_mem(btf, sz);
2882 	if (!t)
2883 		return libbpf_err(-ENOMEM);
2884 
2885 	value_off = btf__add_str(btf, value);
2886 	if (value_off < 0)
2887 		return value_off;
2888 
2889 	t->name_off = value_off;
2890 	t->info = btf_type_info(BTF_KIND_DECL_TAG, 0, kflag);
2891 	t->type = ref_type_id;
2892 	btf_decl_tag(t)->component_idx = component_idx;
2893 
2894 	return btf_commit_type(btf, sz);
2895 }
2896 
2897 /*
2898  * Append new BTF_KIND_DECL_TAG type with:
2899  *   - *value* - non-empty/non-NULL string;
2900  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2901  *   - *component_idx* - -1 for tagging reference type, otherwise struct/union
2902  *     member or function argument index;
2903  * Returns:
2904  *   - >0, type ID of newly added BTF type;
2905  *   - <0, on error.
2906  */
2907 int btf__add_decl_tag(struct btf *btf, const char *value, int ref_type_id,
2908 		      int component_idx)
2909 {
2910 	return btf_add_decl_tag(btf, value, ref_type_id, component_idx, 0);
2911 }
2912 
2913 /*
2914  * Append new BTF_KIND_DECL_TAG type with:
2915  *   - *value* - non-empty/non-NULL string;
2916  *   - *ref_type_id* - referenced type ID, it might not exist yet;
2917  *   - *component_idx* - -1 for tagging reference type, otherwise struct/union
2918  *     member or function argument index;
2919  * Set info->kflag to 1, indicating this tag is an __attribute__
2920  * Returns:
2921  *   - >0, type ID of newly added BTF type;
2922  *   - <0, on error.
2923  */
2924 int btf__add_decl_attr(struct btf *btf, const char *value, int ref_type_id,
2925 		       int component_idx)
2926 {
2927 	return btf_add_decl_tag(btf, value, ref_type_id, component_idx, 1);
2928 }
2929 
2930 struct btf_ext_sec_info_param {
2931 	__u32 off;
2932 	__u32 len;
2933 	__u32 min_rec_size;
2934 	struct btf_ext_info *ext_info;
2935 	const char *desc;
2936 };
2937 
2938 /*
2939  * Parse a single info subsection of the BTF.ext info data:
2940  *  - validate subsection structure and elements
2941  *  - save info subsection start and sizing details in struct btf_ext
2942  *  - endian-independent operation, for calling before byte-swapping
2943  */
2944 static int btf_ext_parse_sec_info(struct btf_ext *btf_ext,
2945 				  struct btf_ext_sec_info_param *ext_sec,
2946 				  bool is_native)
2947 {
2948 	const struct btf_ext_info_sec *sinfo;
2949 	struct btf_ext_info *ext_info;
2950 	__u32 info_left, record_size;
2951 	size_t sec_cnt = 0;
2952 	void *info;
2953 
2954 	if (ext_sec->len == 0)
2955 		return 0;
2956 
2957 	if (ext_sec->off & 0x03) {
2958 		pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
2959 		     ext_sec->desc);
2960 		return -EINVAL;
2961 	}
2962 
2963 	/* The start of the info sec (including the __u32 record_size). */
2964 	info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
2965 	info_left = ext_sec->len;
2966 
2967 	if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
2968 		pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
2969 			 ext_sec->desc, ext_sec->off, ext_sec->len);
2970 		return -EINVAL;
2971 	}
2972 
2973 	/* At least a record size */
2974 	if (info_left < sizeof(__u32)) {
2975 		pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
2976 		return -EINVAL;
2977 	}
2978 
2979 	/* The record size needs to meet either the minimum standard or, when
2980 	 * handling non-native endianness data, the exact standard so as
2981 	 * to allow safe byte-swapping.
2982 	 */
2983 	record_size = is_native ? *(__u32 *)info : bswap_32(*(__u32 *)info);
2984 	if (record_size < ext_sec->min_rec_size ||
2985 	    (!is_native && record_size != ext_sec->min_rec_size) ||
2986 	    record_size & 0x03) {
2987 		pr_debug("%s section in .BTF.ext has invalid record size %u\n",
2988 			 ext_sec->desc, record_size);
2989 		return -EINVAL;
2990 	}
2991 
2992 	sinfo = info + sizeof(__u32);
2993 	info_left -= sizeof(__u32);
2994 
2995 	/* If no records, return failure now so .BTF.ext won't be used. */
2996 	if (!info_left) {
2997 		pr_debug("%s section in .BTF.ext has no records\n", ext_sec->desc);
2998 		return -EINVAL;
2999 	}
3000 
3001 	while (info_left) {
3002 		unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
3003 		__u64 total_record_size;
3004 		__u32 num_records;
3005 
3006 		if (info_left < sec_hdrlen) {
3007 			pr_debug("%s section header is not found in .BTF.ext\n",
3008 			     ext_sec->desc);
3009 			return -EINVAL;
3010 		}
3011 
3012 		num_records = is_native ? sinfo->num_info : bswap_32(sinfo->num_info);
3013 		if (num_records == 0) {
3014 			pr_debug("%s section has incorrect num_records in .BTF.ext\n",
3015 			     ext_sec->desc);
3016 			return -EINVAL;
3017 		}
3018 
3019 		total_record_size = sec_hdrlen + (__u64)num_records * record_size;
3020 		if (info_left < total_record_size) {
3021 			pr_debug("%s section has incorrect num_records in .BTF.ext\n",
3022 			     ext_sec->desc);
3023 			return -EINVAL;
3024 		}
3025 
3026 		info_left -= total_record_size;
3027 		sinfo = (void *)sinfo + total_record_size;
3028 		sec_cnt++;
3029 	}
3030 
3031 	ext_info = ext_sec->ext_info;
3032 	ext_info->len = ext_sec->len - sizeof(__u32);
3033 	ext_info->rec_size = record_size;
3034 	ext_info->info = info + sizeof(__u32);
3035 	ext_info->sec_cnt = sec_cnt;
3036 
3037 	return 0;
3038 }
3039 
3040 /* Parse all info secs in the BTF.ext info data */
3041 static int btf_ext_parse_info(struct btf_ext *btf_ext, bool is_native)
3042 {
3043 	struct btf_ext_sec_info_param func_info = {
3044 		.off = btf_ext->hdr->func_info_off,
3045 		.len = btf_ext->hdr->func_info_len,
3046 		.min_rec_size = sizeof(struct bpf_func_info_min),
3047 		.ext_info = &btf_ext->func_info,
3048 		.desc = "func_info"
3049 	};
3050 	struct btf_ext_sec_info_param line_info = {
3051 		.off = btf_ext->hdr->line_info_off,
3052 		.len = btf_ext->hdr->line_info_len,
3053 		.min_rec_size = sizeof(struct bpf_line_info_min),
3054 		.ext_info = &btf_ext->line_info,
3055 		.desc = "line_info",
3056 	};
3057 	struct btf_ext_sec_info_param core_relo = {
3058 		.min_rec_size = sizeof(struct bpf_core_relo),
3059 		.ext_info = &btf_ext->core_relo_info,
3060 		.desc = "core_relo",
3061 	};
3062 	int err;
3063 
3064 	err = btf_ext_parse_sec_info(btf_ext, &func_info, is_native);
3065 	if (err)
3066 		return err;
3067 
3068 	err = btf_ext_parse_sec_info(btf_ext, &line_info, is_native);
3069 	if (err)
3070 		return err;
3071 
3072 	if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
3073 		return 0; /* skip core relos parsing */
3074 
3075 	core_relo.off = btf_ext->hdr->core_relo_off;
3076 	core_relo.len = btf_ext->hdr->core_relo_len;
3077 	err = btf_ext_parse_sec_info(btf_ext, &core_relo, is_native);
3078 	if (err)
3079 		return err;
3080 
3081 	return 0;
3082 }
3083 
3084 /* Swap byte-order of BTF.ext header with any endianness */
3085 static void btf_ext_bswap_hdr(struct btf_ext_header *h)
3086 {
3087 	bool is_native = h->magic == BTF_MAGIC;
3088 	__u32 hdr_len;
3089 
3090 	hdr_len = is_native ? h->hdr_len : bswap_32(h->hdr_len);
3091 
3092 	h->magic = bswap_16(h->magic);
3093 	h->hdr_len = bswap_32(h->hdr_len);
3094 	h->func_info_off = bswap_32(h->func_info_off);
3095 	h->func_info_len = bswap_32(h->func_info_len);
3096 	h->line_info_off = bswap_32(h->line_info_off);
3097 	h->line_info_len = bswap_32(h->line_info_len);
3098 
3099 	if (hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
3100 		return;
3101 
3102 	h->core_relo_off = bswap_32(h->core_relo_off);
3103 	h->core_relo_len = bswap_32(h->core_relo_len);
3104 }
3105 
3106 /* Swap byte-order of generic info subsection */
3107 static void btf_ext_bswap_info_sec(void *info, __u32 len, bool is_native,
3108 				   info_rec_bswap_fn bswap_fn)
3109 {
3110 	struct btf_ext_info_sec *sec;
3111 	__u32 info_left, rec_size, *rs;
3112 
3113 	if (len == 0)
3114 		return;
3115 
3116 	rs = info;				/* info record size */
3117 	rec_size = is_native ? *rs : bswap_32(*rs);
3118 	*rs = bswap_32(*rs);
3119 
3120 	sec = info + sizeof(__u32);		/* info sec #1 */
3121 	info_left = len - sizeof(__u32);
3122 	while (info_left) {
3123 		unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
3124 		__u32 i, num_recs;
3125 		void *p;
3126 
3127 		num_recs = is_native ? sec->num_info : bswap_32(sec->num_info);
3128 		sec->sec_name_off = bswap_32(sec->sec_name_off);
3129 		sec->num_info = bswap_32(sec->num_info);
3130 		p = sec->data;			/* info rec #1 */
3131 		for (i = 0; i < num_recs; i++, p += rec_size)
3132 			bswap_fn(p);
3133 		sec = p;
3134 		info_left -= sec_hdrlen + (__u64)rec_size * num_recs;
3135 	}
3136 }
3137 
3138 /*
3139  * Swap byte-order of all info data in a BTF.ext section
3140  *  - requires BTF.ext hdr in native endianness
3141  */
3142 static void btf_ext_bswap_info(struct btf_ext *btf_ext, void *data)
3143 {
3144 	const bool is_native = btf_ext->swapped_endian;
3145 	const struct btf_ext_header *h = data;
3146 	void *info;
3147 
3148 	/* Swap func_info subsection byte-order */
3149 	info = data + h->hdr_len + h->func_info_off;
3150 	btf_ext_bswap_info_sec(info, h->func_info_len, is_native,
3151 			       (info_rec_bswap_fn)bpf_func_info_bswap);
3152 
3153 	/* Swap line_info subsection byte-order */
3154 	info = data + h->hdr_len + h->line_info_off;
3155 	btf_ext_bswap_info_sec(info, h->line_info_len, is_native,
3156 			       (info_rec_bswap_fn)bpf_line_info_bswap);
3157 
3158 	/* Swap core_relo subsection byte-order (if present) */
3159 	if (h->hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
3160 		return;
3161 
3162 	info = data + h->hdr_len + h->core_relo_off;
3163 	btf_ext_bswap_info_sec(info, h->core_relo_len, is_native,
3164 			       (info_rec_bswap_fn)bpf_core_relo_bswap);
3165 }
3166 
3167 /* Parse hdr data and info sections: check and convert to native endianness */
3168 static int btf_ext_parse(struct btf_ext *btf_ext)
3169 {
3170 	__u32 hdr_len, data_size = btf_ext->data_size;
3171 	struct btf_ext_header *hdr = btf_ext->hdr;
3172 	bool swapped_endian = false;
3173 	int err;
3174 
3175 	if (data_size < offsetofend(struct btf_ext_header, hdr_len)) {
3176 		pr_debug("BTF.ext header too short\n");
3177 		return -EINVAL;
3178 	}
3179 
3180 	hdr_len = hdr->hdr_len;
3181 	if (hdr->magic == bswap_16(BTF_MAGIC)) {
3182 		swapped_endian = true;
3183 		hdr_len = bswap_32(hdr_len);
3184 	} else if (hdr->magic != BTF_MAGIC) {
3185 		pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
3186 		return -EINVAL;
3187 	}
3188 
3189 	/* Ensure known version of structs, current BTF_VERSION == 1 */
3190 	if (hdr->version != 1) {
3191 		pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
3192 		return -ENOTSUP;
3193 	}
3194 
3195 	if (hdr->flags) {
3196 		pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
3197 		return -ENOTSUP;
3198 	}
3199 
3200 	if (data_size < hdr_len) {
3201 		pr_debug("BTF.ext header not found\n");
3202 		return -EINVAL;
3203 	} else if (data_size == hdr_len) {
3204 		pr_debug("BTF.ext has no data\n");
3205 		return -EINVAL;
3206 	}
3207 
3208 	/* Verify mandatory hdr info details present */
3209 	if (hdr_len < offsetofend(struct btf_ext_header, line_info_len)) {
3210 		pr_warn("BTF.ext header missing func_info, line_info\n");
3211 		return -EINVAL;
3212 	}
3213 
3214 	/* Keep hdr native byte-order in memory for introspection */
3215 	if (swapped_endian)
3216 		btf_ext_bswap_hdr(btf_ext->hdr);
3217 
3218 	/* Validate info subsections and cache key metadata */
3219 	err = btf_ext_parse_info(btf_ext, !swapped_endian);
3220 	if (err)
3221 		return err;
3222 
3223 	/* Keep infos native byte-order in memory for introspection */
3224 	if (swapped_endian)
3225 		btf_ext_bswap_info(btf_ext, btf_ext->data);
3226 
3227 	/*
3228 	 * Set btf_ext->swapped_endian only after all header and info data has
3229 	 * been swapped, helping bswap functions determine if their data are
3230 	 * in native byte-order when called.
3231 	 */
3232 	btf_ext->swapped_endian = swapped_endian;
3233 	return 0;
3234 }
3235 
3236 void btf_ext__free(struct btf_ext *btf_ext)
3237 {
3238 	if (IS_ERR_OR_NULL(btf_ext))
3239 		return;
3240 	free(btf_ext->func_info.sec_idxs);
3241 	free(btf_ext->line_info.sec_idxs);
3242 	free(btf_ext->core_relo_info.sec_idxs);
3243 	free(btf_ext->data);
3244 	free(btf_ext->data_swapped);
3245 	free(btf_ext);
3246 }
3247 
3248 struct btf_ext *btf_ext__new(const __u8 *data, __u32 size)
3249 {
3250 	struct btf_ext *btf_ext;
3251 	int err;
3252 
3253 	btf_ext = calloc(1, sizeof(struct btf_ext));
3254 	if (!btf_ext)
3255 		return libbpf_err_ptr(-ENOMEM);
3256 
3257 	btf_ext->data_size = size;
3258 	btf_ext->data = malloc(size);
3259 	if (!btf_ext->data) {
3260 		err = -ENOMEM;
3261 		goto done;
3262 	}
3263 	memcpy(btf_ext->data, data, size);
3264 
3265 	err = btf_ext_parse(btf_ext);
3266 
3267 done:
3268 	if (err) {
3269 		btf_ext__free(btf_ext);
3270 		return libbpf_err_ptr(err);
3271 	}
3272 
3273 	return btf_ext;
3274 }
3275 
3276 static void *btf_ext_raw_data(const struct btf_ext *btf_ext_ro, bool swap_endian)
3277 {
3278 	struct btf_ext *btf_ext = (struct btf_ext *)btf_ext_ro;
3279 	const __u32 data_sz = btf_ext->data_size;
3280 	void *data;
3281 
3282 	/* Return native data (always present) or swapped data if present */
3283 	if (!swap_endian)
3284 		return btf_ext->data;
3285 	else if (btf_ext->data_swapped)
3286 		return btf_ext->data_swapped;
3287 
3288 	/* Recreate missing swapped data, then cache and return */
3289 	data = calloc(1, data_sz);
3290 	if (!data)
3291 		return NULL;
3292 	memcpy(data, btf_ext->data, data_sz);
3293 
3294 	btf_ext_bswap_info(btf_ext, data);
3295 	btf_ext_bswap_hdr(data);
3296 	btf_ext->data_swapped = data;
3297 	return data;
3298 }
3299 
3300 const void *btf_ext__raw_data(const struct btf_ext *btf_ext, __u32 *size)
3301 {
3302 	void *data;
3303 
3304 	data = btf_ext_raw_data(btf_ext, btf_ext->swapped_endian);
3305 	if (!data)
3306 		return errno = ENOMEM, NULL;
3307 
3308 	*size = btf_ext->data_size;
3309 	return data;
3310 }
3311 
3312 __attribute__((alias("btf_ext__raw_data")))
3313 const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size);
3314 
3315 enum btf_endianness btf_ext__endianness(const struct btf_ext *btf_ext)
3316 {
3317 	if (is_host_big_endian())
3318 		return btf_ext->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
3319 	else
3320 		return btf_ext->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
3321 }
3322 
3323 int btf_ext__set_endianness(struct btf_ext *btf_ext, enum btf_endianness endian)
3324 {
3325 	if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
3326 		return libbpf_err(-EINVAL);
3327 
3328 	btf_ext->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
3329 
3330 	if (!btf_ext->swapped_endian) {
3331 		free(btf_ext->data_swapped);
3332 		btf_ext->data_swapped = NULL;
3333 	}
3334 	return 0;
3335 }
3336 
3337 struct btf_dedup;
3338 
3339 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts);
3340 static void btf_dedup_free(struct btf_dedup *d);
3341 static int btf_dedup_prep(struct btf_dedup *d);
3342 static int btf_dedup_strings(struct btf_dedup *d);
3343 static int btf_dedup_prim_types(struct btf_dedup *d);
3344 static int btf_dedup_struct_types(struct btf_dedup *d);
3345 static int btf_dedup_ref_types(struct btf_dedup *d);
3346 static int btf_dedup_resolve_fwds(struct btf_dedup *d);
3347 static int btf_dedup_compact_types(struct btf_dedup *d);
3348 static int btf_dedup_remap_types(struct btf_dedup *d);
3349 
3350 /*
3351  * Deduplicate BTF types and strings.
3352  *
3353  * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
3354  * section with all BTF type descriptors and string data. It overwrites that
3355  * memory in-place with deduplicated types and strings without any loss of
3356  * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
3357  * is provided, all the strings referenced from .BTF.ext section are honored
3358  * and updated to point to the right offsets after deduplication.
3359  *
3360  * If function returns with error, type/string data might be garbled and should
3361  * be discarded.
3362  *
3363  * More verbose and detailed description of both problem btf_dedup is solving,
3364  * as well as solution could be found at:
3365  * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
3366  *
3367  * Problem description and justification
3368  * =====================================
3369  *
3370  * BTF type information is typically emitted either as a result of conversion
3371  * from DWARF to BTF or directly by compiler. In both cases, each compilation
3372  * unit contains information about a subset of all the types that are used
3373  * in an application. These subsets are frequently overlapping and contain a lot
3374  * of duplicated information when later concatenated together into a single
3375  * binary. This algorithm ensures that each unique type is represented by single
3376  * BTF type descriptor, greatly reducing resulting size of BTF data.
3377  *
3378  * Compilation unit isolation and subsequent duplication of data is not the only
3379  * problem. The same type hierarchy (e.g., struct and all the type that struct
3380  * references) in different compilation units can be represented in BTF to
3381  * various degrees of completeness (or, rather, incompleteness) due to
3382  * struct/union forward declarations.
3383  *
3384  * Let's take a look at an example, that we'll use to better understand the
3385  * problem (and solution). Suppose we have two compilation units, each using
3386  * same `struct S`, but each of them having incomplete type information about
3387  * struct's fields:
3388  *
3389  * // CU #1:
3390  * struct S;
3391  * struct A {
3392  *	int a;
3393  *	struct A* self;
3394  *	struct S* parent;
3395  * };
3396  * struct B;
3397  * struct S {
3398  *	struct A* a_ptr;
3399  *	struct B* b_ptr;
3400  * };
3401  *
3402  * // CU #2:
3403  * struct S;
3404  * struct A;
3405  * struct B {
3406  *	int b;
3407  *	struct B* self;
3408  *	struct S* parent;
3409  * };
3410  * struct S {
3411  *	struct A* a_ptr;
3412  *	struct B* b_ptr;
3413  * };
3414  *
3415  * In case of CU #1, BTF data will know only that `struct B` exist (but no
3416  * more), but will know the complete type information about `struct A`. While
3417  * for CU #2, it will know full type information about `struct B`, but will
3418  * only know about forward declaration of `struct A` (in BTF terms, it will
3419  * have `BTF_KIND_FWD` type descriptor with name `B`).
3420  *
3421  * This compilation unit isolation means that it's possible that there is no
3422  * single CU with complete type information describing structs `S`, `A`, and
3423  * `B`. Also, we might get tons of duplicated and redundant type information.
3424  *
3425  * Additional complication we need to keep in mind comes from the fact that
3426  * types, in general, can form graphs containing cycles, not just DAGs.
3427  *
3428  * While algorithm does deduplication, it also merges and resolves type
3429  * information (unless disabled throught `struct btf_opts`), whenever possible.
3430  * E.g., in the example above with two compilation units having partial type
3431  * information for structs `A` and `B`, the output of algorithm will emit
3432  * a single copy of each BTF type that describes structs `A`, `B`, and `S`
3433  * (as well as type information for `int` and pointers), as if they were defined
3434  * in a single compilation unit as:
3435  *
3436  * struct A {
3437  *	int a;
3438  *	struct A* self;
3439  *	struct S* parent;
3440  * };
3441  * struct B {
3442  *	int b;
3443  *	struct B* self;
3444  *	struct S* parent;
3445  * };
3446  * struct S {
3447  *	struct A* a_ptr;
3448  *	struct B* b_ptr;
3449  * };
3450  *
3451  * Algorithm summary
3452  * =================
3453  *
3454  * Algorithm completes its work in 7 separate passes:
3455  *
3456  * 1. Strings deduplication.
3457  * 2. Primitive types deduplication (int, enum, fwd).
3458  * 3. Struct/union types deduplication.
3459  * 4. Resolve unambiguous forward declarations.
3460  * 5. Reference types deduplication (pointers, typedefs, arrays, funcs, func
3461  *    protos, and const/volatile/restrict modifiers).
3462  * 6. Types compaction.
3463  * 7. Types remapping.
3464  *
3465  * Algorithm determines canonical type descriptor, which is a single
3466  * representative type for each truly unique type. This canonical type is the
3467  * one that will go into final deduplicated BTF type information. For
3468  * struct/unions, it is also the type that algorithm will merge additional type
3469  * information into (while resolving FWDs), as it discovers it from data in
3470  * other CUs. Each input BTF type eventually gets either mapped to itself, if
3471  * that type is canonical, or to some other type, if that type is equivalent
3472  * and was chosen as canonical representative. This mapping is stored in
3473  * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
3474  * FWD type got resolved to.
3475  *
3476  * To facilitate fast discovery of canonical types, we also maintain canonical
3477  * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
3478  * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
3479  * that match that signature. With sufficiently good choice of type signature
3480  * hashing function, we can limit number of canonical types for each unique type
3481  * signature to a very small number, allowing to find canonical type for any
3482  * duplicated type very quickly.
3483  *
3484  * Struct/union deduplication is the most critical part and algorithm for
3485  * deduplicating structs/unions is described in greater details in comments for
3486  * `btf_dedup_is_equiv` function.
3487  */
3488 int btf__dedup(struct btf *btf, const struct btf_dedup_opts *opts)
3489 {
3490 	struct btf_dedup *d;
3491 	int err;
3492 
3493 	if (!OPTS_VALID(opts, btf_dedup_opts))
3494 		return libbpf_err(-EINVAL);
3495 
3496 	d = btf_dedup_new(btf, opts);
3497 	if (IS_ERR(d)) {
3498 		pr_debug("btf_dedup_new failed: %ld\n", PTR_ERR(d));
3499 		return libbpf_err(-EINVAL);
3500 	}
3501 
3502 	if (btf_ensure_modifiable(btf)) {
3503 		err = -ENOMEM;
3504 		goto done;
3505 	}
3506 
3507 	err = btf_dedup_prep(d);
3508 	if (err) {
3509 		pr_debug("btf_dedup_prep failed: %s\n", errstr(err));
3510 		goto done;
3511 	}
3512 	err = btf_dedup_strings(d);
3513 	if (err < 0) {
3514 		pr_debug("btf_dedup_strings failed: %s\n", errstr(err));
3515 		goto done;
3516 	}
3517 	err = btf_dedup_prim_types(d);
3518 	if (err < 0) {
3519 		pr_debug("btf_dedup_prim_types failed: %s\n", errstr(err));
3520 		goto done;
3521 	}
3522 	err = btf_dedup_struct_types(d);
3523 	if (err < 0) {
3524 		pr_debug("btf_dedup_struct_types failed: %s\n", errstr(err));
3525 		goto done;
3526 	}
3527 	err = btf_dedup_resolve_fwds(d);
3528 	if (err < 0) {
3529 		pr_debug("btf_dedup_resolve_fwds failed: %s\n", errstr(err));
3530 		goto done;
3531 	}
3532 	err = btf_dedup_ref_types(d);
3533 	if (err < 0) {
3534 		pr_debug("btf_dedup_ref_types failed: %s\n", errstr(err));
3535 		goto done;
3536 	}
3537 	err = btf_dedup_compact_types(d);
3538 	if (err < 0) {
3539 		pr_debug("btf_dedup_compact_types failed: %s\n", errstr(err));
3540 		goto done;
3541 	}
3542 	err = btf_dedup_remap_types(d);
3543 	if (err < 0) {
3544 		pr_debug("btf_dedup_remap_types failed: %s\n", errstr(err));
3545 		goto done;
3546 	}
3547 
3548 done:
3549 	btf_dedup_free(d);
3550 	return libbpf_err(err);
3551 }
3552 
3553 #define BTF_UNPROCESSED_ID ((__u32)-1)
3554 #define BTF_IN_PROGRESS_ID ((__u32)-2)
3555 
3556 struct btf_dedup {
3557 	/* .BTF section to be deduped in-place */
3558 	struct btf *btf;
3559 	/*
3560 	 * Optional .BTF.ext section. When provided, any strings referenced
3561 	 * from it will be taken into account when deduping strings
3562 	 */
3563 	struct btf_ext *btf_ext;
3564 	/*
3565 	 * This is a map from any type's signature hash to a list of possible
3566 	 * canonical representative type candidates. Hash collisions are
3567 	 * ignored, so even types of various kinds can share same list of
3568 	 * candidates, which is fine because we rely on subsequent
3569 	 * btf_xxx_equal() checks to authoritatively verify type equality.
3570 	 */
3571 	struct hashmap *dedup_table;
3572 	/* Canonical types map */
3573 	__u32 *map;
3574 	/* Hypothetical mapping, used during type graph equivalence checks */
3575 	__u32 *hypot_map;
3576 	__u32 *hypot_list;
3577 	size_t hypot_cnt;
3578 	size_t hypot_cap;
3579 	/* Whether hypothetical mapping, if successful, would need to adjust
3580 	 * already canonicalized types (due to a new forward declaration to
3581 	 * concrete type resolution). In such case, during split BTF dedup
3582 	 * candidate type would still be considered as different, because base
3583 	 * BTF is considered to be immutable.
3584 	 */
3585 	bool hypot_adjust_canon;
3586 	/* Various option modifying behavior of algorithm */
3587 	struct btf_dedup_opts opts;
3588 	/* temporary strings deduplication state */
3589 	struct strset *strs_set;
3590 };
3591 
3592 static unsigned long hash_combine(unsigned long h, unsigned long value)
3593 {
3594 	return h * 31 + value;
3595 }
3596 
3597 #define for_each_dedup_cand(d, node, hash) \
3598 	hashmap__for_each_key_entry(d->dedup_table, node, hash)
3599 
3600 static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
3601 {
3602 	return hashmap__append(d->dedup_table, hash, type_id);
3603 }
3604 
3605 static int btf_dedup_hypot_map_add(struct btf_dedup *d,
3606 				   __u32 from_id, __u32 to_id)
3607 {
3608 	if (d->hypot_cnt == d->hypot_cap) {
3609 		__u32 *new_list;
3610 
3611 		d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
3612 		new_list = libbpf_reallocarray(d->hypot_list, d->hypot_cap, sizeof(__u32));
3613 		if (!new_list)
3614 			return -ENOMEM;
3615 		d->hypot_list = new_list;
3616 	}
3617 	d->hypot_list[d->hypot_cnt++] = from_id;
3618 	d->hypot_map[from_id] = to_id;
3619 	return 0;
3620 }
3621 
3622 static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
3623 {
3624 	int i;
3625 
3626 	for (i = 0; i < d->hypot_cnt; i++)
3627 		d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
3628 	d->hypot_cnt = 0;
3629 	d->hypot_adjust_canon = false;
3630 }
3631 
3632 static void btf_dedup_free(struct btf_dedup *d)
3633 {
3634 	hashmap__free(d->dedup_table);
3635 	d->dedup_table = NULL;
3636 
3637 	free(d->map);
3638 	d->map = NULL;
3639 
3640 	free(d->hypot_map);
3641 	d->hypot_map = NULL;
3642 
3643 	free(d->hypot_list);
3644 	d->hypot_list = NULL;
3645 
3646 	free(d);
3647 }
3648 
3649 static size_t btf_dedup_identity_hash_fn(long key, void *ctx)
3650 {
3651 	return key;
3652 }
3653 
3654 static size_t btf_dedup_collision_hash_fn(long key, void *ctx)
3655 {
3656 	return 0;
3657 }
3658 
3659 static bool btf_dedup_equal_fn(long k1, long k2, void *ctx)
3660 {
3661 	return k1 == k2;
3662 }
3663 
3664 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts)
3665 {
3666 	struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
3667 	hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
3668 	int i, err = 0, type_cnt;
3669 
3670 	if (!d)
3671 		return ERR_PTR(-ENOMEM);
3672 
3673 	if (OPTS_GET(opts, force_collisions, false))
3674 		hash_fn = btf_dedup_collision_hash_fn;
3675 
3676 	d->btf = btf;
3677 	d->btf_ext = OPTS_GET(opts, btf_ext, NULL);
3678 
3679 	d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
3680 	if (IS_ERR(d->dedup_table)) {
3681 		err = PTR_ERR(d->dedup_table);
3682 		d->dedup_table = NULL;
3683 		goto done;
3684 	}
3685 
3686 	type_cnt = btf__type_cnt(btf);
3687 	d->map = malloc(sizeof(__u32) * type_cnt);
3688 	if (!d->map) {
3689 		err = -ENOMEM;
3690 		goto done;
3691 	}
3692 	/* special BTF "void" type is made canonical immediately */
3693 	d->map[0] = 0;
3694 	for (i = 1; i < type_cnt; i++) {
3695 		struct btf_type *t = btf_type_by_id(d->btf, i);
3696 
3697 		/* VAR and DATASEC are never deduped and are self-canonical */
3698 		if (btf_is_var(t) || btf_is_datasec(t))
3699 			d->map[i] = i;
3700 		else
3701 			d->map[i] = BTF_UNPROCESSED_ID;
3702 	}
3703 
3704 	d->hypot_map = malloc(sizeof(__u32) * type_cnt);
3705 	if (!d->hypot_map) {
3706 		err = -ENOMEM;
3707 		goto done;
3708 	}
3709 	for (i = 0; i < type_cnt; i++)
3710 		d->hypot_map[i] = BTF_UNPROCESSED_ID;
3711 
3712 done:
3713 	if (err) {
3714 		btf_dedup_free(d);
3715 		return ERR_PTR(err);
3716 	}
3717 
3718 	return d;
3719 }
3720 
3721 /*
3722  * Iterate over all possible places in .BTF and .BTF.ext that can reference
3723  * string and pass pointer to it to a provided callback `fn`.
3724  */
3725 static int btf_for_each_str_off(struct btf_dedup *d, str_off_visit_fn fn, void *ctx)
3726 {
3727 	int i, r;
3728 
3729 	for (i = 0; i < d->btf->nr_types; i++) {
3730 		struct btf_field_iter it;
3731 		struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
3732 		__u32 *str_off;
3733 
3734 		r = btf_field_iter_init(&it, t, BTF_FIELD_ITER_STRS);
3735 		if (r)
3736 			return r;
3737 
3738 		while ((str_off = btf_field_iter_next(&it))) {
3739 			r = fn(str_off, ctx);
3740 			if (r)
3741 				return r;
3742 		}
3743 	}
3744 
3745 	if (!d->btf_ext)
3746 		return 0;
3747 
3748 	r = btf_ext_visit_str_offs(d->btf_ext, fn, ctx);
3749 	if (r)
3750 		return r;
3751 
3752 	return 0;
3753 }
3754 
3755 static int strs_dedup_remap_str_off(__u32 *str_off_ptr, void *ctx)
3756 {
3757 	struct btf_dedup *d = ctx;
3758 	__u32 str_off = *str_off_ptr;
3759 	const char *s;
3760 	int off, err;
3761 
3762 	/* don't touch empty string or string in main BTF */
3763 	if (str_off == 0 || str_off < d->btf->start_str_off)
3764 		return 0;
3765 
3766 	s = btf__str_by_offset(d->btf, str_off);
3767 	if (d->btf->base_btf) {
3768 		err = btf__find_str(d->btf->base_btf, s);
3769 		if (err >= 0) {
3770 			*str_off_ptr = err;
3771 			return 0;
3772 		}
3773 		if (err != -ENOENT)
3774 			return err;
3775 	}
3776 
3777 	off = strset__add_str(d->strs_set, s);
3778 	if (off < 0)
3779 		return off;
3780 
3781 	*str_off_ptr = d->btf->start_str_off + off;
3782 	return 0;
3783 }
3784 
3785 /*
3786  * Dedup string and filter out those that are not referenced from either .BTF
3787  * or .BTF.ext (if provided) sections.
3788  *
3789  * This is done by building index of all strings in BTF's string section,
3790  * then iterating over all entities that can reference strings (e.g., type
3791  * names, struct field names, .BTF.ext line info, etc) and marking corresponding
3792  * strings as used. After that all used strings are deduped and compacted into
3793  * sequential blob of memory and new offsets are calculated. Then all the string
3794  * references are iterated again and rewritten using new offsets.
3795  */
3796 static int btf_dedup_strings(struct btf_dedup *d)
3797 {
3798 	int err;
3799 
3800 	if (d->btf->strs_deduped)
3801 		return 0;
3802 
3803 	d->strs_set = strset__new(BTF_MAX_STR_OFFSET, NULL, 0);
3804 	if (IS_ERR(d->strs_set)) {
3805 		err = PTR_ERR(d->strs_set);
3806 		goto err_out;
3807 	}
3808 
3809 	if (!d->btf->base_btf) {
3810 		/* insert empty string; we won't be looking it up during strings
3811 		 * dedup, but it's good to have it for generic BTF string lookups
3812 		 */
3813 		err = strset__add_str(d->strs_set, "");
3814 		if (err < 0)
3815 			goto err_out;
3816 	}
3817 
3818 	/* remap string offsets */
3819 	err = btf_for_each_str_off(d, strs_dedup_remap_str_off, d);
3820 	if (err)
3821 		goto err_out;
3822 
3823 	/* replace BTF string data and hash with deduped ones */
3824 	strset__free(d->btf->strs_set);
3825 	d->btf->hdr->str_len = strset__data_size(d->strs_set);
3826 	d->btf->strs_set = d->strs_set;
3827 	d->strs_set = NULL;
3828 	d->btf->strs_deduped = true;
3829 	return 0;
3830 
3831 err_out:
3832 	strset__free(d->strs_set);
3833 	d->strs_set = NULL;
3834 
3835 	return err;
3836 }
3837 
3838 static long btf_hash_common(struct btf_type *t)
3839 {
3840 	long h;
3841 
3842 	h = hash_combine(0, t->name_off);
3843 	h = hash_combine(h, t->info);
3844 	h = hash_combine(h, t->size);
3845 	return h;
3846 }
3847 
3848 static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
3849 {
3850 	return t1->name_off == t2->name_off &&
3851 	       t1->info == t2->info &&
3852 	       t1->size == t2->size;
3853 }
3854 
3855 /* Calculate type signature hash of INT or TAG. */
3856 static long btf_hash_int_decl_tag(struct btf_type *t)
3857 {
3858 	__u32 info = *(__u32 *)(t + 1);
3859 	long h;
3860 
3861 	h = btf_hash_common(t);
3862 	h = hash_combine(h, info);
3863 	return h;
3864 }
3865 
3866 /* Check structural equality of two INTs or TAGs. */
3867 static bool btf_equal_int_tag(struct btf_type *t1, struct btf_type *t2)
3868 {
3869 	__u32 info1, info2;
3870 
3871 	if (!btf_equal_common(t1, t2))
3872 		return false;
3873 	info1 = *(__u32 *)(t1 + 1);
3874 	info2 = *(__u32 *)(t2 + 1);
3875 	return info1 == info2;
3876 }
3877 
3878 /* Calculate type signature hash of ENUM/ENUM64. */
3879 static long btf_hash_enum(struct btf_type *t)
3880 {
3881 	long h;
3882 
3883 	/* don't hash vlen, enum members and size to support enum fwd resolving */
3884 	h = hash_combine(0, t->name_off);
3885 	return h;
3886 }
3887 
3888 static bool btf_equal_enum_members(struct btf_type *t1, struct btf_type *t2)
3889 {
3890 	const struct btf_enum *m1, *m2;
3891 	__u16 vlen;
3892 	int i;
3893 
3894 	vlen = btf_vlen(t1);
3895 	m1 = btf_enum(t1);
3896 	m2 = btf_enum(t2);
3897 	for (i = 0; i < vlen; i++) {
3898 		if (m1->name_off != m2->name_off || m1->val != m2->val)
3899 			return false;
3900 		m1++;
3901 		m2++;
3902 	}
3903 	return true;
3904 }
3905 
3906 static bool btf_equal_enum64_members(struct btf_type *t1, struct btf_type *t2)
3907 {
3908 	const struct btf_enum64 *m1, *m2;
3909 	__u16 vlen;
3910 	int i;
3911 
3912 	vlen = btf_vlen(t1);
3913 	m1 = btf_enum64(t1);
3914 	m2 = btf_enum64(t2);
3915 	for (i = 0; i < vlen; i++) {
3916 		if (m1->name_off != m2->name_off || m1->val_lo32 != m2->val_lo32 ||
3917 		    m1->val_hi32 != m2->val_hi32)
3918 			return false;
3919 		m1++;
3920 		m2++;
3921 	}
3922 	return true;
3923 }
3924 
3925 /* Check structural equality of two ENUMs or ENUM64s. */
3926 static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
3927 {
3928 	if (!btf_equal_common(t1, t2))
3929 		return false;
3930 
3931 	/* t1 & t2 kinds are identical because of btf_equal_common */
3932 	if (btf_kind(t1) == BTF_KIND_ENUM)
3933 		return btf_equal_enum_members(t1, t2);
3934 	else
3935 		return btf_equal_enum64_members(t1, t2);
3936 }
3937 
3938 static inline bool btf_is_enum_fwd(struct btf_type *t)
3939 {
3940 	return btf_is_any_enum(t) && btf_vlen(t) == 0;
3941 }
3942 
3943 static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
3944 {
3945 	if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3946 		return btf_equal_enum(t1, t2);
3947 	/* At this point either t1 or t2 or both are forward declarations, thus:
3948 	 * - skip comparing vlen because it is zero for forward declarations;
3949 	 * - skip comparing size to allow enum forward declarations
3950 	 *   to be compatible with enum64 full declarations;
3951 	 * - skip comparing kind for the same reason.
3952 	 */
3953 	return t1->name_off == t2->name_off &&
3954 	       btf_is_any_enum(t1) && btf_is_any_enum(t2);
3955 }
3956 
3957 /*
3958  * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
3959  * as referenced type IDs equivalence is established separately during type
3960  * graph equivalence check algorithm.
3961  */
3962 static long btf_hash_struct(struct btf_type *t)
3963 {
3964 	const struct btf_member *member = btf_members(t);
3965 	__u32 vlen = btf_vlen(t);
3966 	long h = btf_hash_common(t);
3967 	int i;
3968 
3969 	for (i = 0; i < vlen; i++) {
3970 		h = hash_combine(h, member->name_off);
3971 		h = hash_combine(h, member->offset);
3972 		/* no hashing of referenced type ID, it can be unresolved yet */
3973 		member++;
3974 	}
3975 	return h;
3976 }
3977 
3978 /*
3979  * Check structural compatibility of two STRUCTs/UNIONs, ignoring referenced
3980  * type IDs. This check is performed during type graph equivalence check and
3981  * referenced types equivalence is checked separately.
3982  */
3983 static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
3984 {
3985 	const struct btf_member *m1, *m2;
3986 	__u16 vlen;
3987 	int i;
3988 
3989 	if (!btf_equal_common(t1, t2))
3990 		return false;
3991 
3992 	vlen = btf_vlen(t1);
3993 	m1 = btf_members(t1);
3994 	m2 = btf_members(t2);
3995 	for (i = 0; i < vlen; i++) {
3996 		if (m1->name_off != m2->name_off || m1->offset != m2->offset)
3997 			return false;
3998 		m1++;
3999 		m2++;
4000 	}
4001 	return true;
4002 }
4003 
4004 /*
4005  * Calculate type signature hash of ARRAY, including referenced type IDs,
4006  * under assumption that they were already resolved to canonical type IDs and
4007  * are not going to change.
4008  */
4009 static long btf_hash_array(struct btf_type *t)
4010 {
4011 	const struct btf_array *info = btf_array(t);
4012 	long h = btf_hash_common(t);
4013 
4014 	h = hash_combine(h, info->type);
4015 	h = hash_combine(h, info->index_type);
4016 	h = hash_combine(h, info->nelems);
4017 	return h;
4018 }
4019 
4020 /*
4021  * Check exact equality of two ARRAYs, taking into account referenced
4022  * type IDs, under assumption that they were already resolved to canonical
4023  * type IDs and are not going to change.
4024  * This function is called during reference types deduplication to compare
4025  * ARRAY to potential canonical representative.
4026  */
4027 static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
4028 {
4029 	const struct btf_array *info1, *info2;
4030 
4031 	if (!btf_equal_common(t1, t2))
4032 		return false;
4033 
4034 	info1 = btf_array(t1);
4035 	info2 = btf_array(t2);
4036 	return info1->type == info2->type &&
4037 	       info1->index_type == info2->index_type &&
4038 	       info1->nelems == info2->nelems;
4039 }
4040 
4041 /*
4042  * Check structural compatibility of two ARRAYs, ignoring referenced type
4043  * IDs. This check is performed during type graph equivalence check and
4044  * referenced types equivalence is checked separately.
4045  */
4046 static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
4047 {
4048 	if (!btf_equal_common(t1, t2))
4049 		return false;
4050 
4051 	return btf_array(t1)->nelems == btf_array(t2)->nelems;
4052 }
4053 
4054 /*
4055  * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
4056  * under assumption that they were already resolved to canonical type IDs and
4057  * are not going to change.
4058  */
4059 static long btf_hash_fnproto(struct btf_type *t)
4060 {
4061 	const struct btf_param *member = btf_params(t);
4062 	__u16 vlen = btf_vlen(t);
4063 	long h = btf_hash_common(t);
4064 	int i;
4065 
4066 	for (i = 0; i < vlen; i++) {
4067 		h = hash_combine(h, member->name_off);
4068 		h = hash_combine(h, member->type);
4069 		member++;
4070 	}
4071 	return h;
4072 }
4073 
4074 /*
4075  * Check exact equality of two FUNC_PROTOs, taking into account referenced
4076  * type IDs, under assumption that they were already resolved to canonical
4077  * type IDs and are not going to change.
4078  * This function is called during reference types deduplication to compare
4079  * FUNC_PROTO to potential canonical representative.
4080  */
4081 static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
4082 {
4083 	const struct btf_param *m1, *m2;
4084 	__u16 vlen;
4085 	int i;
4086 
4087 	if (!btf_equal_common(t1, t2))
4088 		return false;
4089 
4090 	vlen = btf_vlen(t1);
4091 	m1 = btf_params(t1);
4092 	m2 = btf_params(t2);
4093 	for (i = 0; i < vlen; i++) {
4094 		if (m1->name_off != m2->name_off || m1->type != m2->type)
4095 			return false;
4096 		m1++;
4097 		m2++;
4098 	}
4099 	return true;
4100 }
4101 
4102 /*
4103  * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
4104  * IDs. This check is performed during type graph equivalence check and
4105  * referenced types equivalence is checked separately.
4106  */
4107 static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
4108 {
4109 	const struct btf_param *m1, *m2;
4110 	__u16 vlen;
4111 	int i;
4112 
4113 	/* skip return type ID */
4114 	if (t1->name_off != t2->name_off || t1->info != t2->info)
4115 		return false;
4116 
4117 	vlen = btf_vlen(t1);
4118 	m1 = btf_params(t1);
4119 	m2 = btf_params(t2);
4120 	for (i = 0; i < vlen; i++) {
4121 		if (m1->name_off != m2->name_off)
4122 			return false;
4123 		m1++;
4124 		m2++;
4125 	}
4126 	return true;
4127 }
4128 
4129 /* Prepare split BTF for deduplication by calculating hashes of base BTF's
4130  * types and initializing the rest of the state (canonical type mapping) for
4131  * the fixed base BTF part.
4132  */
4133 static int btf_dedup_prep(struct btf_dedup *d)
4134 {
4135 	struct btf_type *t;
4136 	int type_id;
4137 	long h;
4138 
4139 	if (!d->btf->base_btf)
4140 		return 0;
4141 
4142 	for (type_id = 1; type_id < d->btf->start_id; type_id++) {
4143 		t = btf_type_by_id(d->btf, type_id);
4144 
4145 		/* all base BTF types are self-canonical by definition */
4146 		d->map[type_id] = type_id;
4147 
4148 		switch (btf_kind(t)) {
4149 		case BTF_KIND_VAR:
4150 		case BTF_KIND_DATASEC:
4151 			/* VAR and DATASEC are never hash/deduplicated */
4152 			continue;
4153 		case BTF_KIND_CONST:
4154 		case BTF_KIND_VOLATILE:
4155 		case BTF_KIND_RESTRICT:
4156 		case BTF_KIND_PTR:
4157 		case BTF_KIND_FWD:
4158 		case BTF_KIND_TYPEDEF:
4159 		case BTF_KIND_FUNC:
4160 		case BTF_KIND_FLOAT:
4161 		case BTF_KIND_TYPE_TAG:
4162 			h = btf_hash_common(t);
4163 			break;
4164 		case BTF_KIND_INT:
4165 		case BTF_KIND_DECL_TAG:
4166 			h = btf_hash_int_decl_tag(t);
4167 			break;
4168 		case BTF_KIND_ENUM:
4169 		case BTF_KIND_ENUM64:
4170 			h = btf_hash_enum(t);
4171 			break;
4172 		case BTF_KIND_STRUCT:
4173 		case BTF_KIND_UNION:
4174 			h = btf_hash_struct(t);
4175 			break;
4176 		case BTF_KIND_ARRAY:
4177 			h = btf_hash_array(t);
4178 			break;
4179 		case BTF_KIND_FUNC_PROTO:
4180 			h = btf_hash_fnproto(t);
4181 			break;
4182 		default:
4183 			pr_debug("unknown kind %d for type [%d]\n", btf_kind(t), type_id);
4184 			return -EINVAL;
4185 		}
4186 		if (btf_dedup_table_add(d, h, type_id))
4187 			return -ENOMEM;
4188 	}
4189 
4190 	return 0;
4191 }
4192 
4193 /*
4194  * Deduplicate primitive types, that can't reference other types, by calculating
4195  * their type signature hash and comparing them with any possible canonical
4196  * candidate. If no canonical candidate matches, type itself is marked as
4197  * canonical and is added into `btf_dedup->dedup_table` as another candidate.
4198  */
4199 static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
4200 {
4201 	struct btf_type *t = btf_type_by_id(d->btf, type_id);
4202 	struct hashmap_entry *hash_entry;
4203 	struct btf_type *cand;
4204 	/* if we don't find equivalent type, then we are canonical */
4205 	__u32 new_id = type_id;
4206 	__u32 cand_id;
4207 	long h;
4208 
4209 	switch (btf_kind(t)) {
4210 	case BTF_KIND_CONST:
4211 	case BTF_KIND_VOLATILE:
4212 	case BTF_KIND_RESTRICT:
4213 	case BTF_KIND_PTR:
4214 	case BTF_KIND_TYPEDEF:
4215 	case BTF_KIND_ARRAY:
4216 	case BTF_KIND_STRUCT:
4217 	case BTF_KIND_UNION:
4218 	case BTF_KIND_FUNC:
4219 	case BTF_KIND_FUNC_PROTO:
4220 	case BTF_KIND_VAR:
4221 	case BTF_KIND_DATASEC:
4222 	case BTF_KIND_DECL_TAG:
4223 	case BTF_KIND_TYPE_TAG:
4224 		return 0;
4225 
4226 	case BTF_KIND_INT:
4227 		h = btf_hash_int_decl_tag(t);
4228 		for_each_dedup_cand(d, hash_entry, h) {
4229 			cand_id = hash_entry->value;
4230 			cand = btf_type_by_id(d->btf, cand_id);
4231 			if (btf_equal_int_tag(t, cand)) {
4232 				new_id = cand_id;
4233 				break;
4234 			}
4235 		}
4236 		break;
4237 
4238 	case BTF_KIND_ENUM:
4239 	case BTF_KIND_ENUM64:
4240 		h = btf_hash_enum(t);
4241 		for_each_dedup_cand(d, hash_entry, h) {
4242 			cand_id = hash_entry->value;
4243 			cand = btf_type_by_id(d->btf, cand_id);
4244 			if (btf_equal_enum(t, cand)) {
4245 				new_id = cand_id;
4246 				break;
4247 			}
4248 			if (btf_compat_enum(t, cand)) {
4249 				if (btf_is_enum_fwd(t)) {
4250 					/* resolve fwd to full enum */
4251 					new_id = cand_id;
4252 					break;
4253 				}
4254 				/* resolve canonical enum fwd to full enum */
4255 				d->map[cand_id] = type_id;
4256 			}
4257 		}
4258 		break;
4259 
4260 	case BTF_KIND_FWD:
4261 	case BTF_KIND_FLOAT:
4262 		h = btf_hash_common(t);
4263 		for_each_dedup_cand(d, hash_entry, h) {
4264 			cand_id = hash_entry->value;
4265 			cand = btf_type_by_id(d->btf, cand_id);
4266 			if (btf_equal_common(t, cand)) {
4267 				new_id = cand_id;
4268 				break;
4269 			}
4270 		}
4271 		break;
4272 
4273 	default:
4274 		return -EINVAL;
4275 	}
4276 
4277 	d->map[type_id] = new_id;
4278 	if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4279 		return -ENOMEM;
4280 
4281 	return 0;
4282 }
4283 
4284 static int btf_dedup_prim_types(struct btf_dedup *d)
4285 {
4286 	int i, err;
4287 
4288 	for (i = 0; i < d->btf->nr_types; i++) {
4289 		err = btf_dedup_prim_type(d, d->btf->start_id + i);
4290 		if (err)
4291 			return err;
4292 	}
4293 	return 0;
4294 }
4295 
4296 /*
4297  * Check whether type is already mapped into canonical one (could be to itself).
4298  */
4299 static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
4300 {
4301 	return d->map[type_id] <= BTF_MAX_NR_TYPES;
4302 }
4303 
4304 /*
4305  * Resolve type ID into its canonical type ID, if any; otherwise return original
4306  * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
4307  * STRUCT/UNION link and resolve it into canonical type ID as well.
4308  */
4309 static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
4310 {
4311 	while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
4312 		type_id = d->map[type_id];
4313 	return type_id;
4314 }
4315 
4316 /*
4317  * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
4318  * type ID.
4319  */
4320 static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
4321 {
4322 	__u32 orig_type_id = type_id;
4323 
4324 	if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
4325 		return type_id;
4326 
4327 	while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
4328 		type_id = d->map[type_id];
4329 
4330 	if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
4331 		return type_id;
4332 
4333 	return orig_type_id;
4334 }
4335 
4336 
4337 static inline __u16 btf_fwd_kind(struct btf_type *t)
4338 {
4339 	return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
4340 }
4341 
4342 /* Check if given two types are identical ARRAY definitions */
4343 static bool btf_dedup_identical_arrays(struct btf_dedup *d, __u32 id1, __u32 id2)
4344 {
4345 	struct btf_type *t1, *t2;
4346 
4347 	t1 = btf_type_by_id(d->btf, id1);
4348 	t2 = btf_type_by_id(d->btf, id2);
4349 	if (!btf_is_array(t1) || !btf_is_array(t2))
4350 		return false;
4351 
4352 	return btf_equal_array(t1, t2);
4353 }
4354 
4355 /* Check if given two types are identical STRUCT/UNION definitions */
4356 static bool btf_dedup_identical_structs(struct btf_dedup *d, __u32 id1, __u32 id2)
4357 {
4358 	const struct btf_member *m1, *m2;
4359 	struct btf_type *t1, *t2;
4360 	int n, i;
4361 
4362 	t1 = btf_type_by_id(d->btf, id1);
4363 	t2 = btf_type_by_id(d->btf, id2);
4364 
4365 	if (!btf_is_composite(t1) || btf_kind(t1) != btf_kind(t2))
4366 		return false;
4367 
4368 	if (!btf_shallow_equal_struct(t1, t2))
4369 		return false;
4370 
4371 	m1 = btf_members(t1);
4372 	m2 = btf_members(t2);
4373 	for (i = 0, n = btf_vlen(t1); i < n; i++, m1++, m2++) {
4374 		if (m1->type != m2->type &&
4375 		    !btf_dedup_identical_arrays(d, m1->type, m2->type) &&
4376 		    !btf_dedup_identical_structs(d, m1->type, m2->type))
4377 			return false;
4378 	}
4379 	return true;
4380 }
4381 
4382 /*
4383  * Check equivalence of BTF type graph formed by candidate struct/union (we'll
4384  * call it "candidate graph" in this description for brevity) to a type graph
4385  * formed by (potential) canonical struct/union ("canonical graph" for brevity
4386  * here, though keep in mind that not all types in canonical graph are
4387  * necessarily canonical representatives themselves, some of them might be
4388  * duplicates or its uniqueness might not have been established yet).
4389  * Returns:
4390  *  - >0, if type graphs are equivalent;
4391  *  -  0, if not equivalent;
4392  *  - <0, on error.
4393  *
4394  * Algorithm performs side-by-side DFS traversal of both type graphs and checks
4395  * equivalence of BTF types at each step. If at any point BTF types in candidate
4396  * and canonical graphs are not compatible structurally, whole graphs are
4397  * incompatible. If types are structurally equivalent (i.e., all information
4398  * except referenced type IDs is exactly the same), a mapping from `canon_id` to
4399  * a `cand_id` is recoded in hypothetical mapping (`btf_dedup->hypot_map`).
4400  * If a type references other types, then those referenced types are checked
4401  * for equivalence recursively.
4402  *
4403  * During DFS traversal, if we find that for current `canon_id` type we
4404  * already have some mapping in hypothetical map, we check for two possible
4405  * situations:
4406  *   - `canon_id` is mapped to exactly the same type as `cand_id`. This will
4407  *     happen when type graphs have cycles. In this case we assume those two
4408  *     types are equivalent.
4409  *   - `canon_id` is mapped to different type. This is contradiction in our
4410  *     hypothetical mapping, because same graph in canonical graph corresponds
4411  *     to two different types in candidate graph, which for equivalent type
4412  *     graphs shouldn't happen. This condition terminates equivalence check
4413  *     with negative result.
4414  *
4415  * If type graphs traversal exhausts types to check and find no contradiction,
4416  * then type graphs are equivalent.
4417  *
4418  * When checking types for equivalence, there is one special case: FWD types.
4419  * If FWD type resolution is allowed and one of the types (either from canonical
4420  * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
4421  * flag) and their names match, hypothetical mapping is updated to point from
4422  * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
4423  * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
4424  *
4425  * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
4426  * if there are two exactly named (or anonymous) structs/unions that are
4427  * compatible structurally, one of which has FWD field, while other is concrete
4428  * STRUCT/UNION, but according to C sources they are different structs/unions
4429  * that are referencing different types with the same name. This is extremely
4430  * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
4431  * this logic is causing problems.
4432  *
4433  * Doing FWD resolution means that both candidate and/or canonical graphs can
4434  * consists of portions of the graph that come from multiple compilation units.
4435  * This is due to the fact that types within single compilation unit are always
4436  * deduplicated and FWDs are already resolved, if referenced struct/union
4437  * definition is available. So, if we had unresolved FWD and found corresponding
4438  * STRUCT/UNION, they will be from different compilation units. This
4439  * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
4440  * type graph will likely have at least two different BTF types that describe
4441  * same type (e.g., most probably there will be two different BTF types for the
4442  * same 'int' primitive type) and could even have "overlapping" parts of type
4443  * graph that describe same subset of types.
4444  *
4445  * This in turn means that our assumption that each type in canonical graph
4446  * must correspond to exactly one type in candidate graph might not hold
4447  * anymore and will make it harder to detect contradictions using hypothetical
4448  * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
4449  * resolution only in canonical graph. FWDs in candidate graphs are never
4450  * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
4451  * that can occur:
4452  *   - Both types in canonical and candidate graphs are FWDs. If they are
4453  *     structurally equivalent, then they can either be both resolved to the
4454  *     same STRUCT/UNION or not resolved at all. In both cases they are
4455  *     equivalent and there is no need to resolve FWD on candidate side.
4456  *   - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
4457  *     so nothing to resolve as well, algorithm will check equivalence anyway.
4458  *   - Type in canonical graph is FWD, while type in candidate is concrete
4459  *     STRUCT/UNION. In this case candidate graph comes from single compilation
4460  *     unit, so there is exactly one BTF type for each unique C type. After
4461  *     resolving FWD into STRUCT/UNION, there might be more than one BTF type
4462  *     in canonical graph mapping to single BTF type in candidate graph, but
4463  *     because hypothetical mapping maps from canonical to candidate types, it's
4464  *     alright, and we still maintain the property of having single `canon_id`
4465  *     mapping to single `cand_id` (there could be two different `canon_id`
4466  *     mapped to the same `cand_id`, but it's not contradictory).
4467  *   - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
4468  *     graph is FWD. In this case we are just going to check compatibility of
4469  *     STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
4470  *     assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
4471  *     a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
4472  *     turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
4473  *     canonical graph.
4474  */
4475 static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
4476 			      __u32 canon_id)
4477 {
4478 	struct btf_type *cand_type;
4479 	struct btf_type *canon_type;
4480 	__u32 hypot_type_id;
4481 	__u16 cand_kind;
4482 	__u16 canon_kind;
4483 	int i, eq;
4484 
4485 	/* if both resolve to the same canonical, they must be equivalent */
4486 	if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
4487 		return 1;
4488 
4489 	canon_id = resolve_fwd_id(d, canon_id);
4490 
4491 	hypot_type_id = d->hypot_map[canon_id];
4492 	if (hypot_type_id <= BTF_MAX_NR_TYPES) {
4493 		if (hypot_type_id == cand_id)
4494 			return 1;
4495 		/* In some cases compiler will generate different DWARF types
4496 		 * for *identical* array type definitions and use them for
4497 		 * different fields within the *same* struct. This breaks type
4498 		 * equivalence check, which makes an assumption that candidate
4499 		 * types sub-graph has a consistent and deduped-by-compiler
4500 		 * types within a single CU. So work around that by explicitly
4501 		 * allowing identical array types here.
4502 		 */
4503 		if (btf_dedup_identical_arrays(d, hypot_type_id, cand_id))
4504 			return 1;
4505 		/* It turns out that similar situation can happen with
4506 		 * struct/union sometimes, sigh... Handle the case where
4507 		 * structs/unions are exactly the same, down to the referenced
4508 		 * type IDs. Anything more complicated (e.g., if referenced
4509 		 * types are different, but equivalent) is *way more*
4510 		 * complicated and requires a many-to-many equivalence mapping.
4511 		 */
4512 		if (btf_dedup_identical_structs(d, hypot_type_id, cand_id))
4513 			return 1;
4514 		return 0;
4515 	}
4516 
4517 	if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
4518 		return -ENOMEM;
4519 
4520 	cand_type = btf_type_by_id(d->btf, cand_id);
4521 	canon_type = btf_type_by_id(d->btf, canon_id);
4522 	cand_kind = btf_kind(cand_type);
4523 	canon_kind = btf_kind(canon_type);
4524 
4525 	if (cand_type->name_off != canon_type->name_off)
4526 		return 0;
4527 
4528 	/* FWD <--> STRUCT/UNION equivalence check, if enabled */
4529 	if ((cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
4530 	    && cand_kind != canon_kind) {
4531 		__u16 real_kind;
4532 		__u16 fwd_kind;
4533 
4534 		if (cand_kind == BTF_KIND_FWD) {
4535 			real_kind = canon_kind;
4536 			fwd_kind = btf_fwd_kind(cand_type);
4537 		} else {
4538 			real_kind = cand_kind;
4539 			fwd_kind = btf_fwd_kind(canon_type);
4540 			/* we'd need to resolve base FWD to STRUCT/UNION */
4541 			if (fwd_kind == real_kind && canon_id < d->btf->start_id)
4542 				d->hypot_adjust_canon = true;
4543 		}
4544 		return fwd_kind == real_kind;
4545 	}
4546 
4547 	if (cand_kind != canon_kind)
4548 		return 0;
4549 
4550 	switch (cand_kind) {
4551 	case BTF_KIND_INT:
4552 		return btf_equal_int_tag(cand_type, canon_type);
4553 
4554 	case BTF_KIND_ENUM:
4555 	case BTF_KIND_ENUM64:
4556 		return btf_compat_enum(cand_type, canon_type);
4557 
4558 	case BTF_KIND_FWD:
4559 	case BTF_KIND_FLOAT:
4560 		return btf_equal_common(cand_type, canon_type);
4561 
4562 	case BTF_KIND_CONST:
4563 	case BTF_KIND_VOLATILE:
4564 	case BTF_KIND_RESTRICT:
4565 	case BTF_KIND_PTR:
4566 	case BTF_KIND_TYPEDEF:
4567 	case BTF_KIND_FUNC:
4568 	case BTF_KIND_TYPE_TAG:
4569 		if (cand_type->info != canon_type->info)
4570 			return 0;
4571 		return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4572 
4573 	case BTF_KIND_ARRAY: {
4574 		const struct btf_array *cand_arr, *canon_arr;
4575 
4576 		if (!btf_compat_array(cand_type, canon_type))
4577 			return 0;
4578 		cand_arr = btf_array(cand_type);
4579 		canon_arr = btf_array(canon_type);
4580 		eq = btf_dedup_is_equiv(d, cand_arr->index_type, canon_arr->index_type);
4581 		if (eq <= 0)
4582 			return eq;
4583 		return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
4584 	}
4585 
4586 	case BTF_KIND_STRUCT:
4587 	case BTF_KIND_UNION: {
4588 		const struct btf_member *cand_m, *canon_m;
4589 		__u16 vlen;
4590 
4591 		if (!btf_shallow_equal_struct(cand_type, canon_type))
4592 			return 0;
4593 		vlen = btf_vlen(cand_type);
4594 		cand_m = btf_members(cand_type);
4595 		canon_m = btf_members(canon_type);
4596 		for (i = 0; i < vlen; i++) {
4597 			eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
4598 			if (eq <= 0)
4599 				return eq;
4600 			cand_m++;
4601 			canon_m++;
4602 		}
4603 
4604 		return 1;
4605 	}
4606 
4607 	case BTF_KIND_FUNC_PROTO: {
4608 		const struct btf_param *cand_p, *canon_p;
4609 		__u16 vlen;
4610 
4611 		if (!btf_compat_fnproto(cand_type, canon_type))
4612 			return 0;
4613 		eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4614 		if (eq <= 0)
4615 			return eq;
4616 		vlen = btf_vlen(cand_type);
4617 		cand_p = btf_params(cand_type);
4618 		canon_p = btf_params(canon_type);
4619 		for (i = 0; i < vlen; i++) {
4620 			eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
4621 			if (eq <= 0)
4622 				return eq;
4623 			cand_p++;
4624 			canon_p++;
4625 		}
4626 		return 1;
4627 	}
4628 
4629 	default:
4630 		return -EINVAL;
4631 	}
4632 	return 0;
4633 }
4634 
4635 /*
4636  * Use hypothetical mapping, produced by successful type graph equivalence
4637  * check, to augment existing struct/union canonical mapping, where possible.
4638  *
4639  * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
4640  * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
4641  * it doesn't matter if FWD type was part of canonical graph or candidate one,
4642  * we are recording the mapping anyway. As opposed to carefulness required
4643  * for struct/union correspondence mapping (described below), for FWD resolution
4644  * it's not important, as by the time that FWD type (reference type) will be
4645  * deduplicated all structs/unions will be deduped already anyway.
4646  *
4647  * Recording STRUCT/UNION mapping is purely a performance optimization and is
4648  * not required for correctness. It needs to be done carefully to ensure that
4649  * struct/union from candidate's type graph is not mapped into corresponding
4650  * struct/union from canonical type graph that itself hasn't been resolved into
4651  * canonical representative. The only guarantee we have is that canonical
4652  * struct/union was determined as canonical and that won't change. But any
4653  * types referenced through that struct/union fields could have been not yet
4654  * resolved, so in case like that it's too early to establish any kind of
4655  * correspondence between structs/unions.
4656  *
4657  * No canonical correspondence is derived for primitive types (they are already
4658  * deduplicated completely already anyway) or reference types (they rely on
4659  * stability of struct/union canonical relationship for equivalence checks).
4660  */
4661 static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
4662 {
4663 	__u32 canon_type_id, targ_type_id;
4664 	__u16 t_kind, c_kind;
4665 	__u32 t_id, c_id;
4666 	int i;
4667 
4668 	for (i = 0; i < d->hypot_cnt; i++) {
4669 		canon_type_id = d->hypot_list[i];
4670 		targ_type_id = d->hypot_map[canon_type_id];
4671 		t_id = resolve_type_id(d, targ_type_id);
4672 		c_id = resolve_type_id(d, canon_type_id);
4673 		t_kind = btf_kind(btf__type_by_id(d->btf, t_id));
4674 		c_kind = btf_kind(btf__type_by_id(d->btf, c_id));
4675 		/*
4676 		 * Resolve FWD into STRUCT/UNION.
4677 		 * It's ok to resolve FWD into STRUCT/UNION that's not yet
4678 		 * mapped to canonical representative (as opposed to
4679 		 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
4680 		 * eventually that struct is going to be mapped and all resolved
4681 		 * FWDs will automatically resolve to correct canonical
4682 		 * representative. This will happen before ref type deduping,
4683 		 * which critically depends on stability of these mapping. This
4684 		 * stability is not a requirement for STRUCT/UNION equivalence
4685 		 * checks, though.
4686 		 */
4687 
4688 		/* if it's the split BTF case, we still need to point base FWD
4689 		 * to STRUCT/UNION in a split BTF, because FWDs from split BTF
4690 		 * will be resolved against base FWD. If we don't point base
4691 		 * canonical FWD to the resolved STRUCT/UNION, then all the
4692 		 * FWDs in split BTF won't be correctly resolved to a proper
4693 		 * STRUCT/UNION.
4694 		 */
4695 		if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
4696 			d->map[c_id] = t_id;
4697 
4698 		/* if graph equivalence determined that we'd need to adjust
4699 		 * base canonical types, then we need to only point base FWDs
4700 		 * to STRUCTs/UNIONs and do no more modifications. For all
4701 		 * other purposes the type graphs were not equivalent.
4702 		 */
4703 		if (d->hypot_adjust_canon)
4704 			continue;
4705 
4706 		if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
4707 			d->map[t_id] = c_id;
4708 
4709 		if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
4710 		    c_kind != BTF_KIND_FWD &&
4711 		    is_type_mapped(d, c_id) &&
4712 		    !is_type_mapped(d, t_id)) {
4713 			/*
4714 			 * as a perf optimization, we can map struct/union
4715 			 * that's part of type graph we just verified for
4716 			 * equivalence. We can do that for struct/union that has
4717 			 * canonical representative only, though.
4718 			 */
4719 			d->map[t_id] = c_id;
4720 		}
4721 	}
4722 }
4723 
4724 /*
4725  * Deduplicate struct/union types.
4726  *
4727  * For each struct/union type its type signature hash is calculated, taking
4728  * into account type's name, size, number, order and names of fields, but
4729  * ignoring type ID's referenced from fields, because they might not be deduped
4730  * completely until after reference types deduplication phase. This type hash
4731  * is used to iterate over all potential canonical types, sharing same hash.
4732  * For each canonical candidate we check whether type graphs that they form
4733  * (through referenced types in fields and so on) are equivalent using algorithm
4734  * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
4735  * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
4736  * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
4737  * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
4738  * potentially map other structs/unions to their canonical representatives,
4739  * if such relationship hasn't yet been established. This speeds up algorithm
4740  * by eliminating some of the duplicate work.
4741  *
4742  * If no matching canonical representative was found, struct/union is marked
4743  * as canonical for itself and is added into btf_dedup->dedup_table hash map
4744  * for further look ups.
4745  */
4746 static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
4747 {
4748 	struct btf_type *cand_type, *t;
4749 	struct hashmap_entry *hash_entry;
4750 	/* if we don't find equivalent type, then we are canonical */
4751 	__u32 new_id = type_id;
4752 	__u16 kind;
4753 	long h;
4754 
4755 	/* already deduped or is in process of deduping (loop detected) */
4756 	if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4757 		return 0;
4758 
4759 	t = btf_type_by_id(d->btf, type_id);
4760 	kind = btf_kind(t);
4761 
4762 	if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4763 		return 0;
4764 
4765 	h = btf_hash_struct(t);
4766 	for_each_dedup_cand(d, hash_entry, h) {
4767 		__u32 cand_id = hash_entry->value;
4768 		int eq;
4769 
4770 		/*
4771 		 * Even though btf_dedup_is_equiv() checks for
4772 		 * btf_shallow_equal_struct() internally when checking two
4773 		 * structs (unions) for equivalence, we need to guard here
4774 		 * from picking matching FWD type as a dedup candidate.
4775 		 * This can happen due to hash collision. In such case just
4776 		 * relying on btf_dedup_is_equiv() would lead to potentially
4777 		 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
4778 		 * FWD and compatible STRUCT/UNION are considered equivalent.
4779 		 */
4780 		cand_type = btf_type_by_id(d->btf, cand_id);
4781 		if (!btf_shallow_equal_struct(t, cand_type))
4782 			continue;
4783 
4784 		btf_dedup_clear_hypot_map(d);
4785 		eq = btf_dedup_is_equiv(d, type_id, cand_id);
4786 		if (eq < 0)
4787 			return eq;
4788 		if (!eq)
4789 			continue;
4790 		btf_dedup_merge_hypot_map(d);
4791 		if (d->hypot_adjust_canon) /* not really equivalent */
4792 			continue;
4793 		new_id = cand_id;
4794 		break;
4795 	}
4796 
4797 	d->map[type_id] = new_id;
4798 	if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4799 		return -ENOMEM;
4800 
4801 	return 0;
4802 }
4803 
4804 static int btf_dedup_struct_types(struct btf_dedup *d)
4805 {
4806 	int i, err;
4807 
4808 	for (i = 0; i < d->btf->nr_types; i++) {
4809 		err = btf_dedup_struct_type(d, d->btf->start_id + i);
4810 		if (err)
4811 			return err;
4812 	}
4813 	return 0;
4814 }
4815 
4816 /*
4817  * Deduplicate reference type.
4818  *
4819  * Once all primitive and struct/union types got deduplicated, we can easily
4820  * deduplicate all other (reference) BTF types. This is done in two steps:
4821  *
4822  * 1. Resolve all referenced type IDs into their canonical type IDs. This
4823  * resolution can be done either immediately for primitive or struct/union types
4824  * (because they were deduped in previous two phases) or recursively for
4825  * reference types. Recursion will always terminate at either primitive or
4826  * struct/union type, at which point we can "unwind" chain of reference types
4827  * one by one. There is no danger of encountering cycles because in C type
4828  * system the only way to form type cycle is through struct/union, so any chain
4829  * of reference types, even those taking part in a type cycle, will inevitably
4830  * reach struct/union at some point.
4831  *
4832  * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
4833  * becomes "stable", in the sense that no further deduplication will cause
4834  * any changes to it. With that, it's now possible to calculate type's signature
4835  * hash (this time taking into account referenced type IDs) and loop over all
4836  * potential canonical representatives. If no match was found, current type
4837  * will become canonical representative of itself and will be added into
4838  * btf_dedup->dedup_table as another possible canonical representative.
4839  */
4840 static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
4841 {
4842 	struct hashmap_entry *hash_entry;
4843 	__u32 new_id = type_id, cand_id;
4844 	struct btf_type *t, *cand;
4845 	/* if we don't find equivalent type, then we are representative type */
4846 	int ref_type_id;
4847 	long h;
4848 
4849 	if (d->map[type_id] == BTF_IN_PROGRESS_ID)
4850 		return -ELOOP;
4851 	if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4852 		return resolve_type_id(d, type_id);
4853 
4854 	t = btf_type_by_id(d->btf, type_id);
4855 	d->map[type_id] = BTF_IN_PROGRESS_ID;
4856 
4857 	switch (btf_kind(t)) {
4858 	case BTF_KIND_CONST:
4859 	case BTF_KIND_VOLATILE:
4860 	case BTF_KIND_RESTRICT:
4861 	case BTF_KIND_PTR:
4862 	case BTF_KIND_TYPEDEF:
4863 	case BTF_KIND_FUNC:
4864 	case BTF_KIND_TYPE_TAG:
4865 		ref_type_id = btf_dedup_ref_type(d, t->type);
4866 		if (ref_type_id < 0)
4867 			return ref_type_id;
4868 		t->type = ref_type_id;
4869 
4870 		h = btf_hash_common(t);
4871 		for_each_dedup_cand(d, hash_entry, h) {
4872 			cand_id = hash_entry->value;
4873 			cand = btf_type_by_id(d->btf, cand_id);
4874 			if (btf_equal_common(t, cand)) {
4875 				new_id = cand_id;
4876 				break;
4877 			}
4878 		}
4879 		break;
4880 
4881 	case BTF_KIND_DECL_TAG:
4882 		ref_type_id = btf_dedup_ref_type(d, t->type);
4883 		if (ref_type_id < 0)
4884 			return ref_type_id;
4885 		t->type = ref_type_id;
4886 
4887 		h = btf_hash_int_decl_tag(t);
4888 		for_each_dedup_cand(d, hash_entry, h) {
4889 			cand_id = hash_entry->value;
4890 			cand = btf_type_by_id(d->btf, cand_id);
4891 			if (btf_equal_int_tag(t, cand)) {
4892 				new_id = cand_id;
4893 				break;
4894 			}
4895 		}
4896 		break;
4897 
4898 	case BTF_KIND_ARRAY: {
4899 		struct btf_array *info = btf_array(t);
4900 
4901 		ref_type_id = btf_dedup_ref_type(d, info->type);
4902 		if (ref_type_id < 0)
4903 			return ref_type_id;
4904 		info->type = ref_type_id;
4905 
4906 		ref_type_id = btf_dedup_ref_type(d, info->index_type);
4907 		if (ref_type_id < 0)
4908 			return ref_type_id;
4909 		info->index_type = ref_type_id;
4910 
4911 		h = btf_hash_array(t);
4912 		for_each_dedup_cand(d, hash_entry, h) {
4913 			cand_id = hash_entry->value;
4914 			cand = btf_type_by_id(d->btf, cand_id);
4915 			if (btf_equal_array(t, cand)) {
4916 				new_id = cand_id;
4917 				break;
4918 			}
4919 		}
4920 		break;
4921 	}
4922 
4923 	case BTF_KIND_FUNC_PROTO: {
4924 		struct btf_param *param;
4925 		__u16 vlen;
4926 		int i;
4927 
4928 		ref_type_id = btf_dedup_ref_type(d, t->type);
4929 		if (ref_type_id < 0)
4930 			return ref_type_id;
4931 		t->type = ref_type_id;
4932 
4933 		vlen = btf_vlen(t);
4934 		param = btf_params(t);
4935 		for (i = 0; i < vlen; i++) {
4936 			ref_type_id = btf_dedup_ref_type(d, param->type);
4937 			if (ref_type_id < 0)
4938 				return ref_type_id;
4939 			param->type = ref_type_id;
4940 			param++;
4941 		}
4942 
4943 		h = btf_hash_fnproto(t);
4944 		for_each_dedup_cand(d, hash_entry, h) {
4945 			cand_id = hash_entry->value;
4946 			cand = btf_type_by_id(d->btf, cand_id);
4947 			if (btf_equal_fnproto(t, cand)) {
4948 				new_id = cand_id;
4949 				break;
4950 			}
4951 		}
4952 		break;
4953 	}
4954 
4955 	default:
4956 		return -EINVAL;
4957 	}
4958 
4959 	d->map[type_id] = new_id;
4960 	if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4961 		return -ENOMEM;
4962 
4963 	return new_id;
4964 }
4965 
4966 static int btf_dedup_ref_types(struct btf_dedup *d)
4967 {
4968 	int i, err;
4969 
4970 	for (i = 0; i < d->btf->nr_types; i++) {
4971 		err = btf_dedup_ref_type(d, d->btf->start_id + i);
4972 		if (err < 0)
4973 			return err;
4974 	}
4975 	/* we won't need d->dedup_table anymore */
4976 	hashmap__free(d->dedup_table);
4977 	d->dedup_table = NULL;
4978 	return 0;
4979 }
4980 
4981 /*
4982  * Collect a map from type names to type ids for all canonical structs
4983  * and unions. If the same name is shared by several canonical types
4984  * use a special value 0 to indicate this fact.
4985  */
4986 static int btf_dedup_fill_unique_names_map(struct btf_dedup *d, struct hashmap *names_map)
4987 {
4988 	__u32 nr_types = btf__type_cnt(d->btf);
4989 	struct btf_type *t;
4990 	__u32 type_id;
4991 	__u16 kind;
4992 	int err;
4993 
4994 	/*
4995 	 * Iterate over base and split module ids in order to get all
4996 	 * available structs in the map.
4997 	 */
4998 	for (type_id = 1; type_id < nr_types; ++type_id) {
4999 		t = btf_type_by_id(d->btf, type_id);
5000 		kind = btf_kind(t);
5001 
5002 		if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
5003 			continue;
5004 
5005 		/* Skip non-canonical types */
5006 		if (type_id != d->map[type_id])
5007 			continue;
5008 
5009 		err = hashmap__add(names_map, t->name_off, type_id);
5010 		if (err == -EEXIST)
5011 			err = hashmap__set(names_map, t->name_off, 0, NULL, NULL);
5012 
5013 		if (err)
5014 			return err;
5015 	}
5016 
5017 	return 0;
5018 }
5019 
5020 static int btf_dedup_resolve_fwd(struct btf_dedup *d, struct hashmap *names_map, __u32 type_id)
5021 {
5022 	struct btf_type *t = btf_type_by_id(d->btf, type_id);
5023 	enum btf_fwd_kind fwd_kind = btf_kflag(t);
5024 	__u16 cand_kind, kind = btf_kind(t);
5025 	struct btf_type *cand_t;
5026 	uintptr_t cand_id;
5027 
5028 	if (kind != BTF_KIND_FWD)
5029 		return 0;
5030 
5031 	/* Skip if this FWD already has a mapping */
5032 	if (type_id != d->map[type_id])
5033 		return 0;
5034 
5035 	if (!hashmap__find(names_map, t->name_off, &cand_id))
5036 		return 0;
5037 
5038 	/* Zero is a special value indicating that name is not unique */
5039 	if (!cand_id)
5040 		return 0;
5041 
5042 	cand_t = btf_type_by_id(d->btf, cand_id);
5043 	cand_kind = btf_kind(cand_t);
5044 	if ((cand_kind == BTF_KIND_STRUCT && fwd_kind != BTF_FWD_STRUCT) ||
5045 	    (cand_kind == BTF_KIND_UNION && fwd_kind != BTF_FWD_UNION))
5046 		return 0;
5047 
5048 	d->map[type_id] = cand_id;
5049 
5050 	return 0;
5051 }
5052 
5053 /*
5054  * Resolve unambiguous forward declarations.
5055  *
5056  * The lion's share of all FWD declarations is resolved during
5057  * `btf_dedup_struct_types` phase when different type graphs are
5058  * compared against each other. However, if in some compilation unit a
5059  * FWD declaration is not a part of a type graph compared against
5060  * another type graph that declaration's canonical type would not be
5061  * changed. Example:
5062  *
5063  * CU #1:
5064  *
5065  * struct foo;
5066  * struct foo *some_global;
5067  *
5068  * CU #2:
5069  *
5070  * struct foo { int u; };
5071  * struct foo *another_global;
5072  *
5073  * After `btf_dedup_struct_types` the BTF looks as follows:
5074  *
5075  * [1] STRUCT 'foo' size=4 vlen=1 ...
5076  * [2] INT 'int' size=4 ...
5077  * [3] PTR '(anon)' type_id=1
5078  * [4] FWD 'foo' fwd_kind=struct
5079  * [5] PTR '(anon)' type_id=4
5080  *
5081  * This pass assumes that such FWD declarations should be mapped to
5082  * structs or unions with identical name in case if the name is not
5083  * ambiguous.
5084  */
5085 static int btf_dedup_resolve_fwds(struct btf_dedup *d)
5086 {
5087 	int i, err;
5088 	struct hashmap *names_map;
5089 
5090 	names_map = hashmap__new(btf_dedup_identity_hash_fn, btf_dedup_equal_fn, NULL);
5091 	if (IS_ERR(names_map))
5092 		return PTR_ERR(names_map);
5093 
5094 	err = btf_dedup_fill_unique_names_map(d, names_map);
5095 	if (err < 0)
5096 		goto exit;
5097 
5098 	for (i = 0; i < d->btf->nr_types; i++) {
5099 		err = btf_dedup_resolve_fwd(d, names_map, d->btf->start_id + i);
5100 		if (err < 0)
5101 			break;
5102 	}
5103 
5104 exit:
5105 	hashmap__free(names_map);
5106 	return err;
5107 }
5108 
5109 /*
5110  * Compact types.
5111  *
5112  * After we established for each type its corresponding canonical representative
5113  * type, we now can eliminate types that are not canonical and leave only
5114  * canonical ones layed out sequentially in memory by copying them over
5115  * duplicates. During compaction btf_dedup->hypot_map array is reused to store
5116  * a map from original type ID to a new compacted type ID, which will be used
5117  * during next phase to "fix up" type IDs, referenced from struct/union and
5118  * reference types.
5119  */
5120 static int btf_dedup_compact_types(struct btf_dedup *d)
5121 {
5122 	__u32 *new_offs;
5123 	__u32 next_type_id = d->btf->start_id;
5124 	const struct btf_type *t;
5125 	void *p;
5126 	int i, id, len;
5127 
5128 	/* we are going to reuse hypot_map to store compaction remapping */
5129 	d->hypot_map[0] = 0;
5130 	/* base BTF types are not renumbered */
5131 	for (id = 1; id < d->btf->start_id; id++)
5132 		d->hypot_map[id] = id;
5133 	for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++)
5134 		d->hypot_map[id] = BTF_UNPROCESSED_ID;
5135 
5136 	p = d->btf->types_data;
5137 
5138 	for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++) {
5139 		if (d->map[id] != id)
5140 			continue;
5141 
5142 		t = btf__type_by_id(d->btf, id);
5143 		len = btf_type_size(t);
5144 		if (len < 0)
5145 			return len;
5146 
5147 		memmove(p, t, len);
5148 		d->hypot_map[id] = next_type_id;
5149 		d->btf->type_offs[next_type_id - d->btf->start_id] = p - d->btf->types_data;
5150 		p += len;
5151 		next_type_id++;
5152 	}
5153 
5154 	/* shrink struct btf's internal types index and update btf_header */
5155 	d->btf->nr_types = next_type_id - d->btf->start_id;
5156 	d->btf->type_offs_cap = d->btf->nr_types;
5157 	d->btf->hdr->type_len = p - d->btf->types_data;
5158 	new_offs = libbpf_reallocarray(d->btf->type_offs, d->btf->type_offs_cap,
5159 				       sizeof(*new_offs));
5160 	if (d->btf->type_offs_cap && !new_offs)
5161 		return -ENOMEM;
5162 	d->btf->type_offs = new_offs;
5163 	d->btf->hdr->str_off = d->btf->hdr->type_len;
5164 	d->btf->raw_size = d->btf->hdr->hdr_len + d->btf->hdr->type_len + d->btf->hdr->str_len;
5165 	return 0;
5166 }
5167 
5168 /*
5169  * Figure out final (deduplicated and compacted) type ID for provided original
5170  * `type_id` by first resolving it into corresponding canonical type ID and
5171  * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
5172  * which is populated during compaction phase.
5173  */
5174 static int btf_dedup_remap_type_id(__u32 *type_id, void *ctx)
5175 {
5176 	struct btf_dedup *d = ctx;
5177 	__u32 resolved_type_id, new_type_id;
5178 
5179 	resolved_type_id = resolve_type_id(d, *type_id);
5180 	new_type_id = d->hypot_map[resolved_type_id];
5181 	if (new_type_id > BTF_MAX_NR_TYPES)
5182 		return -EINVAL;
5183 
5184 	*type_id = new_type_id;
5185 	return 0;
5186 }
5187 
5188 /*
5189  * Remap referenced type IDs into deduped type IDs.
5190  *
5191  * After BTF types are deduplicated and compacted, their final type IDs may
5192  * differ from original ones. The map from original to a corresponding
5193  * deduped type ID is stored in btf_dedup->hypot_map and is populated during
5194  * compaction phase. During remapping phase we are rewriting all type IDs
5195  * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
5196  * their final deduped type IDs.
5197  */
5198 static int btf_dedup_remap_types(struct btf_dedup *d)
5199 {
5200 	int i, r;
5201 
5202 	for (i = 0; i < d->btf->nr_types; i++) {
5203 		struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
5204 		struct btf_field_iter it;
5205 		__u32 *type_id;
5206 
5207 		r = btf_field_iter_init(&it, t, BTF_FIELD_ITER_IDS);
5208 		if (r)
5209 			return r;
5210 
5211 		while ((type_id = btf_field_iter_next(&it))) {
5212 			__u32 resolved_id, new_id;
5213 
5214 			resolved_id = resolve_type_id(d, *type_id);
5215 			new_id = d->hypot_map[resolved_id];
5216 			if (new_id > BTF_MAX_NR_TYPES)
5217 				return -EINVAL;
5218 
5219 			*type_id = new_id;
5220 		}
5221 	}
5222 
5223 	if (!d->btf_ext)
5224 		return 0;
5225 
5226 	r = btf_ext_visit_type_ids(d->btf_ext, btf_dedup_remap_type_id, d);
5227 	if (r)
5228 		return r;
5229 
5230 	return 0;
5231 }
5232 
5233 /*
5234  * Probe few well-known locations for vmlinux kernel image and try to load BTF
5235  * data out of it to use for target BTF.
5236  */
5237 struct btf *btf__load_vmlinux_btf(void)
5238 {
5239 	const char *sysfs_btf_path = "/sys/kernel/btf/vmlinux";
5240 	/* fall back locations, trying to find vmlinux on disk */
5241 	const char *locations[] = {
5242 		"/boot/vmlinux-%1$s",
5243 		"/lib/modules/%1$s/vmlinux-%1$s",
5244 		"/lib/modules/%1$s/build/vmlinux",
5245 		"/usr/lib/modules/%1$s/kernel/vmlinux",
5246 		"/usr/lib/debug/boot/vmlinux-%1$s",
5247 		"/usr/lib/debug/boot/vmlinux-%1$s.debug",
5248 		"/usr/lib/debug/lib/modules/%1$s/vmlinux",
5249 	};
5250 	char path[PATH_MAX + 1];
5251 	struct utsname buf;
5252 	struct btf *btf;
5253 	int i, err;
5254 
5255 	/* is canonical sysfs location accessible? */
5256 	if (faccessat(AT_FDCWD, sysfs_btf_path, F_OK, AT_EACCESS) < 0) {
5257 		pr_warn("kernel BTF is missing at '%s', was CONFIG_DEBUG_INFO_BTF enabled?\n",
5258 			sysfs_btf_path);
5259 	} else {
5260 		btf = btf__parse(sysfs_btf_path, NULL);
5261 		if (!btf) {
5262 			err = -errno;
5263 			pr_warn("failed to read kernel BTF from '%s': %s\n",
5264 				sysfs_btf_path, errstr(err));
5265 			return libbpf_err_ptr(err);
5266 		}
5267 		pr_debug("loaded kernel BTF from '%s'\n", sysfs_btf_path);
5268 		return btf;
5269 	}
5270 
5271 	/* try fallback locations */
5272 	uname(&buf);
5273 	for (i = 0; i < ARRAY_SIZE(locations); i++) {
5274 		snprintf(path, PATH_MAX, locations[i], buf.release);
5275 
5276 		if (faccessat(AT_FDCWD, path, R_OK, AT_EACCESS))
5277 			continue;
5278 
5279 		btf = btf__parse(path, NULL);
5280 		err = libbpf_get_error(btf);
5281 		pr_debug("loading kernel BTF '%s': %s\n", path, errstr(err));
5282 		if (err)
5283 			continue;
5284 
5285 		return btf;
5286 	}
5287 
5288 	pr_warn("failed to find valid kernel BTF\n");
5289 	return libbpf_err_ptr(-ESRCH);
5290 }
5291 
5292 struct btf *libbpf_find_kernel_btf(void) __attribute__((alias("btf__load_vmlinux_btf")));
5293 
5294 struct btf *btf__load_module_btf(const char *module_name, struct btf *vmlinux_btf)
5295 {
5296 	char path[80];
5297 
5298 	snprintf(path, sizeof(path), "/sys/kernel/btf/%s", module_name);
5299 	return btf__parse_split(path, vmlinux_btf);
5300 }
5301 
5302 int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx)
5303 {
5304 	const struct btf_ext_info *seg;
5305 	struct btf_ext_info_sec *sec;
5306 	int i, err;
5307 
5308 	seg = &btf_ext->func_info;
5309 	for_each_btf_ext_sec(seg, sec) {
5310 		struct bpf_func_info_min *rec;
5311 
5312 		for_each_btf_ext_rec(seg, sec, i, rec) {
5313 			err = visit(&rec->type_id, ctx);
5314 			if (err < 0)
5315 				return err;
5316 		}
5317 	}
5318 
5319 	seg = &btf_ext->core_relo_info;
5320 	for_each_btf_ext_sec(seg, sec) {
5321 		struct bpf_core_relo *rec;
5322 
5323 		for_each_btf_ext_rec(seg, sec, i, rec) {
5324 			err = visit(&rec->type_id, ctx);
5325 			if (err < 0)
5326 				return err;
5327 		}
5328 	}
5329 
5330 	return 0;
5331 }
5332 
5333 int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx)
5334 {
5335 	const struct btf_ext_info *seg;
5336 	struct btf_ext_info_sec *sec;
5337 	int i, err;
5338 
5339 	seg = &btf_ext->func_info;
5340 	for_each_btf_ext_sec(seg, sec) {
5341 		err = visit(&sec->sec_name_off, ctx);
5342 		if (err)
5343 			return err;
5344 	}
5345 
5346 	seg = &btf_ext->line_info;
5347 	for_each_btf_ext_sec(seg, sec) {
5348 		struct bpf_line_info_min *rec;
5349 
5350 		err = visit(&sec->sec_name_off, ctx);
5351 		if (err)
5352 			return err;
5353 
5354 		for_each_btf_ext_rec(seg, sec, i, rec) {
5355 			err = visit(&rec->file_name_off, ctx);
5356 			if (err)
5357 				return err;
5358 			err = visit(&rec->line_off, ctx);
5359 			if (err)
5360 				return err;
5361 		}
5362 	}
5363 
5364 	seg = &btf_ext->core_relo_info;
5365 	for_each_btf_ext_sec(seg, sec) {
5366 		struct bpf_core_relo *rec;
5367 
5368 		err = visit(&sec->sec_name_off, ctx);
5369 		if (err)
5370 			return err;
5371 
5372 		for_each_btf_ext_rec(seg, sec, i, rec) {
5373 			err = visit(&rec->access_str_off, ctx);
5374 			if (err)
5375 				return err;
5376 		}
5377 	}
5378 
5379 	return 0;
5380 }
5381 
5382 struct btf_distill {
5383 	struct btf_pipe pipe;
5384 	int *id_map;
5385 	unsigned int split_start_id;
5386 	unsigned int split_start_str;
5387 	int diff_id;
5388 };
5389 
5390 static int btf_add_distilled_type_ids(struct btf_distill *dist, __u32 i)
5391 {
5392 	struct btf_type *split_t = btf_type_by_id(dist->pipe.src, i);
5393 	struct btf_field_iter it;
5394 	__u32 *id;
5395 	int err;
5396 
5397 	err = btf_field_iter_init(&it, split_t, BTF_FIELD_ITER_IDS);
5398 	if (err)
5399 		return err;
5400 	while ((id = btf_field_iter_next(&it))) {
5401 		struct btf_type *base_t;
5402 
5403 		if (!*id)
5404 			continue;
5405 		/* split BTF id, not needed */
5406 		if (*id >= dist->split_start_id)
5407 			continue;
5408 		/* already added ? */
5409 		if (dist->id_map[*id] > 0)
5410 			continue;
5411 
5412 		/* only a subset of base BTF types should be referenced from
5413 		 * split BTF; ensure nothing unexpected is referenced.
5414 		 */
5415 		base_t = btf_type_by_id(dist->pipe.src, *id);
5416 		switch (btf_kind(base_t)) {
5417 		case BTF_KIND_INT:
5418 		case BTF_KIND_FLOAT:
5419 		case BTF_KIND_FWD:
5420 		case BTF_KIND_ARRAY:
5421 		case BTF_KIND_STRUCT:
5422 		case BTF_KIND_UNION:
5423 		case BTF_KIND_TYPEDEF:
5424 		case BTF_KIND_ENUM:
5425 		case BTF_KIND_ENUM64:
5426 		case BTF_KIND_PTR:
5427 		case BTF_KIND_CONST:
5428 		case BTF_KIND_RESTRICT:
5429 		case BTF_KIND_VOLATILE:
5430 		case BTF_KIND_FUNC_PROTO:
5431 		case BTF_KIND_TYPE_TAG:
5432 			dist->id_map[*id] = *id;
5433 			break;
5434 		default:
5435 			pr_warn("unexpected reference to base type[%u] of kind [%u] when creating distilled base BTF.\n",
5436 				*id, btf_kind(base_t));
5437 			return -EINVAL;
5438 		}
5439 		/* If a base type is used, ensure types it refers to are
5440 		 * marked as used also; so for example if we find a PTR to INT
5441 		 * we need both the PTR and INT.
5442 		 *
5443 		 * The only exception is named struct/unions, since distilled
5444 		 * base BTF composite types have no members.
5445 		 */
5446 		if (btf_is_composite(base_t) && base_t->name_off)
5447 			continue;
5448 		err = btf_add_distilled_type_ids(dist, *id);
5449 		if (err)
5450 			return err;
5451 	}
5452 	return 0;
5453 }
5454 
5455 static int btf_add_distilled_types(struct btf_distill *dist)
5456 {
5457 	bool adding_to_base = dist->pipe.dst->start_id == 1;
5458 	int id = btf__type_cnt(dist->pipe.dst);
5459 	struct btf_type *t;
5460 	int i, err = 0;
5461 
5462 
5463 	/* Add types for each of the required references to either distilled
5464 	 * base or split BTF, depending on type characteristics.
5465 	 */
5466 	for (i = 1; i < dist->split_start_id; i++) {
5467 		const char *name;
5468 		int kind;
5469 
5470 		if (!dist->id_map[i])
5471 			continue;
5472 		t = btf_type_by_id(dist->pipe.src, i);
5473 		kind = btf_kind(t);
5474 		name = btf__name_by_offset(dist->pipe.src, t->name_off);
5475 
5476 		switch (kind) {
5477 		case BTF_KIND_INT:
5478 		case BTF_KIND_FLOAT:
5479 		case BTF_KIND_FWD:
5480 			/* Named int, float, fwd are added to base. */
5481 			if (!adding_to_base)
5482 				continue;
5483 			err = btf_add_type(&dist->pipe, t);
5484 			break;
5485 		case BTF_KIND_STRUCT:
5486 		case BTF_KIND_UNION:
5487 			/* Named struct/union are added to base as 0-vlen
5488 			 * struct/union of same size.  Anonymous struct/unions
5489 			 * are added to split BTF as-is.
5490 			 */
5491 			if (adding_to_base) {
5492 				if (!t->name_off)
5493 					continue;
5494 				err = btf_add_composite(dist->pipe.dst, kind, name, t->size);
5495 			} else {
5496 				if (t->name_off)
5497 					continue;
5498 				err = btf_add_type(&dist->pipe, t);
5499 			}
5500 			break;
5501 		case BTF_KIND_ENUM:
5502 		case BTF_KIND_ENUM64:
5503 			/* Named enum[64]s are added to base as a sized
5504 			 * enum; relocation will match with appropriately-named
5505 			 * and sized enum or enum64.
5506 			 *
5507 			 * Anonymous enums are added to split BTF as-is.
5508 			 */
5509 			if (adding_to_base) {
5510 				if (!t->name_off)
5511 					continue;
5512 				err = btf__add_enum(dist->pipe.dst, name, t->size);
5513 			} else {
5514 				if (t->name_off)
5515 					continue;
5516 				err = btf_add_type(&dist->pipe, t);
5517 			}
5518 			break;
5519 		case BTF_KIND_ARRAY:
5520 		case BTF_KIND_TYPEDEF:
5521 		case BTF_KIND_PTR:
5522 		case BTF_KIND_CONST:
5523 		case BTF_KIND_RESTRICT:
5524 		case BTF_KIND_VOLATILE:
5525 		case BTF_KIND_FUNC_PROTO:
5526 		case BTF_KIND_TYPE_TAG:
5527 			/* All other types are added to split BTF. */
5528 			if (adding_to_base)
5529 				continue;
5530 			err = btf_add_type(&dist->pipe, t);
5531 			break;
5532 		default:
5533 			pr_warn("unexpected kind when adding base type '%s'[%u] of kind [%u] to distilled base BTF.\n",
5534 				name, i, kind);
5535 			return -EINVAL;
5536 
5537 		}
5538 		if (err < 0)
5539 			break;
5540 		dist->id_map[i] = id++;
5541 	}
5542 	return err;
5543 }
5544 
5545 /* Split BTF ids without a mapping will be shifted downwards since distilled
5546  * base BTF is smaller than the original base BTF.  For those that have a
5547  * mapping (either to base or updated split BTF), update the id based on
5548  * that mapping.
5549  */
5550 static int btf_update_distilled_type_ids(struct btf_distill *dist, __u32 i)
5551 {
5552 	struct btf_type *t = btf_type_by_id(dist->pipe.dst, i);
5553 	struct btf_field_iter it;
5554 	__u32 *id;
5555 	int err;
5556 
5557 	err = btf_field_iter_init(&it, t, BTF_FIELD_ITER_IDS);
5558 	if (err)
5559 		return err;
5560 	while ((id = btf_field_iter_next(&it))) {
5561 		if (dist->id_map[*id])
5562 			*id = dist->id_map[*id];
5563 		else if (*id >= dist->split_start_id)
5564 			*id -= dist->diff_id;
5565 	}
5566 	return 0;
5567 }
5568 
5569 /* Create updated split BTF with distilled base BTF; distilled base BTF
5570  * consists of BTF information required to clarify the types that split
5571  * BTF refers to, omitting unneeded details.  Specifically it will contain
5572  * base types and memberless definitions of named structs, unions and enumerated
5573  * types. Associated reference types like pointers, arrays and anonymous
5574  * structs, unions and enumerated types will be added to split BTF.
5575  * Size is recorded for named struct/unions to help guide matching to the
5576  * target base BTF during later relocation.
5577  *
5578  * The only case where structs, unions or enumerated types are fully represented
5579  * is when they are anonymous; in such cases, the anonymous type is added to
5580  * split BTF in full.
5581  *
5582  * We return newly-created split BTF where the split BTF refers to a newly-created
5583  * distilled base BTF. Both must be freed separately by the caller.
5584  */
5585 int btf__distill_base(const struct btf *src_btf, struct btf **new_base_btf,
5586 		      struct btf **new_split_btf)
5587 {
5588 	struct btf *new_base = NULL, *new_split = NULL;
5589 	const struct btf *old_base;
5590 	unsigned int n = btf__type_cnt(src_btf);
5591 	struct btf_distill dist = {};
5592 	struct btf_type *t;
5593 	int i, err = 0;
5594 
5595 	/* src BTF must be split BTF. */
5596 	old_base = btf__base_btf(src_btf);
5597 	if (!new_base_btf || !new_split_btf || !old_base)
5598 		return libbpf_err(-EINVAL);
5599 
5600 	new_base = btf__new_empty();
5601 	if (!new_base)
5602 		return libbpf_err(-ENOMEM);
5603 
5604 	btf__set_endianness(new_base, btf__endianness(src_btf));
5605 
5606 	dist.id_map = calloc(n, sizeof(*dist.id_map));
5607 	if (!dist.id_map) {
5608 		err = -ENOMEM;
5609 		goto done;
5610 	}
5611 	dist.pipe.src = src_btf;
5612 	dist.pipe.dst = new_base;
5613 	dist.pipe.str_off_map = hashmap__new(btf_dedup_identity_hash_fn, btf_dedup_equal_fn, NULL);
5614 	if (IS_ERR(dist.pipe.str_off_map)) {
5615 		err = -ENOMEM;
5616 		goto done;
5617 	}
5618 	dist.split_start_id = btf__type_cnt(old_base);
5619 	dist.split_start_str = old_base->hdr->str_len;
5620 
5621 	/* Pass over src split BTF; generate the list of base BTF type ids it
5622 	 * references; these will constitute our distilled BTF set to be
5623 	 * distributed over base and split BTF as appropriate.
5624 	 */
5625 	for (i = src_btf->start_id; i < n; i++) {
5626 		err = btf_add_distilled_type_ids(&dist, i);
5627 		if (err < 0)
5628 			goto done;
5629 	}
5630 	/* Next add types for each of the required references to base BTF and split BTF
5631 	 * in turn.
5632 	 */
5633 	err = btf_add_distilled_types(&dist);
5634 	if (err < 0)
5635 		goto done;
5636 
5637 	/* Create new split BTF with distilled base BTF as its base; the final
5638 	 * state is split BTF with distilled base BTF that represents enough
5639 	 * about its base references to allow it to be relocated with the base
5640 	 * BTF available.
5641 	 */
5642 	new_split = btf__new_empty_split(new_base);
5643 	if (!new_split) {
5644 		err = -errno;
5645 		goto done;
5646 	}
5647 	dist.pipe.dst = new_split;
5648 	/* First add all split types */
5649 	for (i = src_btf->start_id; i < n; i++) {
5650 		t = btf_type_by_id(src_btf, i);
5651 		err = btf_add_type(&dist.pipe, t);
5652 		if (err < 0)
5653 			goto done;
5654 	}
5655 	/* Now add distilled types to split BTF that are not added to base. */
5656 	err = btf_add_distilled_types(&dist);
5657 	if (err < 0)
5658 		goto done;
5659 
5660 	/* All split BTF ids will be shifted downwards since there are less base
5661 	 * BTF ids in distilled base BTF.
5662 	 */
5663 	dist.diff_id = dist.split_start_id - btf__type_cnt(new_base);
5664 
5665 	n = btf__type_cnt(new_split);
5666 	/* Now update base/split BTF ids. */
5667 	for (i = 1; i < n; i++) {
5668 		err = btf_update_distilled_type_ids(&dist, i);
5669 		if (err < 0)
5670 			break;
5671 	}
5672 done:
5673 	free(dist.id_map);
5674 	hashmap__free(dist.pipe.str_off_map);
5675 	if (err) {
5676 		btf__free(new_split);
5677 		btf__free(new_base);
5678 		return libbpf_err(err);
5679 	}
5680 	*new_base_btf = new_base;
5681 	*new_split_btf = new_split;
5682 
5683 	return 0;
5684 }
5685 
5686 const struct btf_header *btf_header(const struct btf *btf)
5687 {
5688 	return btf->hdr;
5689 }
5690 
5691 void btf_set_base_btf(struct btf *btf, const struct btf *base_btf)
5692 {
5693 	btf->base_btf = (struct btf *)base_btf;
5694 	btf->start_id = btf__type_cnt(base_btf);
5695 	btf->start_str_off = base_btf->hdr->str_len;
5696 }
5697 
5698 int btf__relocate(struct btf *btf, const struct btf *base_btf)
5699 {
5700 	int err = btf_relocate(btf, base_btf, NULL);
5701 
5702 	if (!err)
5703 		btf->owns_base = false;
5704 	return libbpf_err(err);
5705 }
5706