1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2 /* Copyright (c) 2018 Facebook */
3
4 #include <byteswap.h>
5 #include <endian.h>
6 #include <stdio.h>
7 #include <stdlib.h>
8 #include <string.h>
9 #include <fcntl.h>
10 #include <unistd.h>
11 #include <errno.h>
12 #include <sys/utsname.h>
13 #include <sys/param.h>
14 #include <sys/stat.h>
15 #include <linux/kernel.h>
16 #include <linux/err.h>
17 #include <linux/btf.h>
18 #include <gelf.h>
19 #include "btf.h"
20 #include "bpf.h"
21 #include "libbpf.h"
22 #include "libbpf_internal.h"
23 #include "hashmap.h"
24 #include "strset.h"
25
26 #define BTF_MAX_NR_TYPES 0x7fffffffU
27 #define BTF_MAX_STR_OFFSET 0x7fffffffU
28
29 static struct btf_type btf_void;
30
31 struct btf {
32 /* raw BTF data in native endianness */
33 void *raw_data;
34 /* raw BTF data in non-native endianness */
35 void *raw_data_swapped;
36 __u32 raw_size;
37 /* whether target endianness differs from the native one */
38 bool swapped_endian;
39
40 /*
41 * When BTF is loaded from an ELF or raw memory it is stored
42 * in a contiguous memory block. The hdr, type_data, and, strs_data
43 * point inside that memory region to their respective parts of BTF
44 * representation:
45 *
46 * +--------------------------------+
47 * | Header | Types | Strings |
48 * +--------------------------------+
49 * ^ ^ ^
50 * | | |
51 * hdr | |
52 * types_data-+ |
53 * strs_data------------+
54 *
55 * If BTF data is later modified, e.g., due to types added or
56 * removed, BTF deduplication performed, etc, this contiguous
57 * representation is broken up into three independently allocated
58 * memory regions to be able to modify them independently.
59 * raw_data is nulled out at that point, but can be later allocated
60 * and cached again if user calls btf__raw_data(), at which point
61 * raw_data will contain a contiguous copy of header, types, and
62 * strings:
63 *
64 * +----------+ +---------+ +-----------+
65 * | Header | | Types | | Strings |
66 * +----------+ +---------+ +-----------+
67 * ^ ^ ^
68 * | | |
69 * hdr | |
70 * types_data----+ |
71 * strset__data(strs_set)-----+
72 *
73 * +----------+---------+-----------+
74 * | Header | Types | Strings |
75 * raw_data----->+----------+---------+-----------+
76 */
77 struct btf_header *hdr;
78
79 void *types_data;
80 size_t types_data_cap; /* used size stored in hdr->type_len */
81
82 /* type ID to `struct btf_type *` lookup index
83 * type_offs[0] corresponds to the first non-VOID type:
84 * - for base BTF it's type [1];
85 * - for split BTF it's the first non-base BTF type.
86 */
87 __u32 *type_offs;
88 size_t type_offs_cap;
89 /* number of types in this BTF instance:
90 * - doesn't include special [0] void type;
91 * - for split BTF counts number of types added on top of base BTF.
92 */
93 __u32 nr_types;
94 /* if not NULL, points to the base BTF on top of which the current
95 * split BTF is based
96 */
97 struct btf *base_btf;
98 /* BTF type ID of the first type in this BTF instance:
99 * - for base BTF it's equal to 1;
100 * - for split BTF it's equal to biggest type ID of base BTF plus 1.
101 */
102 int start_id;
103 /* logical string offset of this BTF instance:
104 * - for base BTF it's equal to 0;
105 * - for split BTF it's equal to total size of base BTF's string section size.
106 */
107 int start_str_off;
108
109 /* only one of strs_data or strs_set can be non-NULL, depending on
110 * whether BTF is in a modifiable state (strs_set is used) or not
111 * (strs_data points inside raw_data)
112 */
113 void *strs_data;
114 /* a set of unique strings */
115 struct strset *strs_set;
116 /* whether strings are already deduplicated */
117 bool strs_deduped;
118
119 /* whether base_btf should be freed in btf_free for this instance */
120 bool owns_base;
121
122 /* BTF object FD, if loaded into kernel */
123 int fd;
124
125 /* Pointer size (in bytes) for a target architecture of this BTF */
126 int ptr_sz;
127 };
128
ptr_to_u64(const void * ptr)129 static inline __u64 ptr_to_u64(const void *ptr)
130 {
131 return (__u64) (unsigned long) ptr;
132 }
133
134 /* Ensure given dynamically allocated memory region pointed to by *data* with
135 * capacity of *cap_cnt* elements each taking *elem_sz* bytes has enough
136 * memory to accommodate *add_cnt* new elements, assuming *cur_cnt* elements
137 * are already used. At most *max_cnt* elements can be ever allocated.
138 * If necessary, memory is reallocated and all existing data is copied over,
139 * new pointer to the memory region is stored at *data, new memory region
140 * capacity (in number of elements) is stored in *cap.
141 * On success, memory pointer to the beginning of unused memory is returned.
142 * On error, NULL is returned.
143 */
libbpf_add_mem(void ** data,size_t * cap_cnt,size_t elem_sz,size_t cur_cnt,size_t max_cnt,size_t add_cnt)144 void *libbpf_add_mem(void **data, size_t *cap_cnt, size_t elem_sz,
145 size_t cur_cnt, size_t max_cnt, size_t add_cnt)
146 {
147 size_t new_cnt;
148 void *new_data;
149
150 if (cur_cnt + add_cnt <= *cap_cnt)
151 return *data + cur_cnt * elem_sz;
152
153 /* requested more than the set limit */
154 if (cur_cnt + add_cnt > max_cnt)
155 return NULL;
156
157 new_cnt = *cap_cnt;
158 new_cnt += new_cnt / 4; /* expand by 25% */
159 if (new_cnt < 16) /* but at least 16 elements */
160 new_cnt = 16;
161 if (new_cnt > max_cnt) /* but not exceeding a set limit */
162 new_cnt = max_cnt;
163 if (new_cnt < cur_cnt + add_cnt) /* also ensure we have enough memory */
164 new_cnt = cur_cnt + add_cnt;
165
166 new_data = libbpf_reallocarray(*data, new_cnt, elem_sz);
167 if (!new_data)
168 return NULL;
169
170 /* zero out newly allocated portion of memory */
171 memset(new_data + (*cap_cnt) * elem_sz, 0, (new_cnt - *cap_cnt) * elem_sz);
172
173 *data = new_data;
174 *cap_cnt = new_cnt;
175 return new_data + cur_cnt * elem_sz;
176 }
177
178 /* Ensure given dynamically allocated memory region has enough allocated space
179 * to accommodate *need_cnt* elements of size *elem_sz* bytes each
180 */
libbpf_ensure_mem(void ** data,size_t * cap_cnt,size_t elem_sz,size_t need_cnt)181 int libbpf_ensure_mem(void **data, size_t *cap_cnt, size_t elem_sz, size_t need_cnt)
182 {
183 void *p;
184
185 if (need_cnt <= *cap_cnt)
186 return 0;
187
188 p = libbpf_add_mem(data, cap_cnt, elem_sz, *cap_cnt, SIZE_MAX, need_cnt - *cap_cnt);
189 if (!p)
190 return -ENOMEM;
191
192 return 0;
193 }
194
btf_add_type_offs_mem(struct btf * btf,size_t add_cnt)195 static void *btf_add_type_offs_mem(struct btf *btf, size_t add_cnt)
196 {
197 return libbpf_add_mem((void **)&btf->type_offs, &btf->type_offs_cap, sizeof(__u32),
198 btf->nr_types, BTF_MAX_NR_TYPES, add_cnt);
199 }
200
btf_add_type_idx_entry(struct btf * btf,__u32 type_off)201 static int btf_add_type_idx_entry(struct btf *btf, __u32 type_off)
202 {
203 __u32 *p;
204
205 p = btf_add_type_offs_mem(btf, 1);
206 if (!p)
207 return -ENOMEM;
208
209 *p = type_off;
210 return 0;
211 }
212
btf_bswap_hdr(struct btf_header * h)213 static void btf_bswap_hdr(struct btf_header *h)
214 {
215 h->magic = bswap_16(h->magic);
216 h->hdr_len = bswap_32(h->hdr_len);
217 h->type_off = bswap_32(h->type_off);
218 h->type_len = bswap_32(h->type_len);
219 h->str_off = bswap_32(h->str_off);
220 h->str_len = bswap_32(h->str_len);
221 }
222
btf_parse_hdr(struct btf * btf)223 static int btf_parse_hdr(struct btf *btf)
224 {
225 struct btf_header *hdr = btf->hdr;
226 __u32 meta_left;
227
228 if (btf->raw_size < sizeof(struct btf_header)) {
229 pr_debug("BTF header not found\n");
230 return -EINVAL;
231 }
232
233 if (hdr->magic == bswap_16(BTF_MAGIC)) {
234 btf->swapped_endian = true;
235 if (bswap_32(hdr->hdr_len) != sizeof(struct btf_header)) {
236 pr_warn("Can't load BTF with non-native endianness due to unsupported header length %u\n",
237 bswap_32(hdr->hdr_len));
238 return -ENOTSUP;
239 }
240 btf_bswap_hdr(hdr);
241 } else if (hdr->magic != BTF_MAGIC) {
242 pr_debug("Invalid BTF magic: %x\n", hdr->magic);
243 return -EINVAL;
244 }
245
246 if (btf->raw_size < hdr->hdr_len) {
247 pr_debug("BTF header len %u larger than data size %u\n",
248 hdr->hdr_len, btf->raw_size);
249 return -EINVAL;
250 }
251
252 meta_left = btf->raw_size - hdr->hdr_len;
253 if (meta_left < (long long)hdr->str_off + hdr->str_len) {
254 pr_debug("Invalid BTF total size: %u\n", btf->raw_size);
255 return -EINVAL;
256 }
257
258 if ((long long)hdr->type_off + hdr->type_len > hdr->str_off) {
259 pr_debug("Invalid BTF data sections layout: type data at %u + %u, strings data at %u + %u\n",
260 hdr->type_off, hdr->type_len, hdr->str_off, hdr->str_len);
261 return -EINVAL;
262 }
263
264 if (hdr->type_off % 4) {
265 pr_debug("BTF type section is not aligned to 4 bytes\n");
266 return -EINVAL;
267 }
268
269 return 0;
270 }
271
btf_parse_str_sec(struct btf * btf)272 static int btf_parse_str_sec(struct btf *btf)
273 {
274 const struct btf_header *hdr = btf->hdr;
275 const char *start = btf->strs_data;
276 const char *end = start + btf->hdr->str_len;
277
278 if (btf->base_btf && hdr->str_len == 0)
279 return 0;
280 if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET || end[-1]) {
281 pr_debug("Invalid BTF string section\n");
282 return -EINVAL;
283 }
284 if (!btf->base_btf && start[0]) {
285 pr_debug("Invalid BTF string section\n");
286 return -EINVAL;
287 }
288 return 0;
289 }
290
btf_type_size(const struct btf_type * t)291 static int btf_type_size(const struct btf_type *t)
292 {
293 const int base_size = sizeof(struct btf_type);
294 __u16 vlen = btf_vlen(t);
295
296 switch (btf_kind(t)) {
297 case BTF_KIND_FWD:
298 case BTF_KIND_CONST:
299 case BTF_KIND_VOLATILE:
300 case BTF_KIND_RESTRICT:
301 case BTF_KIND_PTR:
302 case BTF_KIND_TYPEDEF:
303 case BTF_KIND_FUNC:
304 case BTF_KIND_FLOAT:
305 case BTF_KIND_TYPE_TAG:
306 return base_size;
307 case BTF_KIND_INT:
308 return base_size + sizeof(__u32);
309 case BTF_KIND_ENUM:
310 return base_size + vlen * sizeof(struct btf_enum);
311 case BTF_KIND_ENUM64:
312 return base_size + vlen * sizeof(struct btf_enum64);
313 case BTF_KIND_ARRAY:
314 return base_size + sizeof(struct btf_array);
315 case BTF_KIND_STRUCT:
316 case BTF_KIND_UNION:
317 return base_size + vlen * sizeof(struct btf_member);
318 case BTF_KIND_FUNC_PROTO:
319 return base_size + vlen * sizeof(struct btf_param);
320 case BTF_KIND_VAR:
321 return base_size + sizeof(struct btf_var);
322 case BTF_KIND_DATASEC:
323 return base_size + vlen * sizeof(struct btf_var_secinfo);
324 case BTF_KIND_DECL_TAG:
325 return base_size + sizeof(struct btf_decl_tag);
326 default:
327 pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
328 return -EINVAL;
329 }
330 }
331
btf_bswap_type_base(struct btf_type * t)332 static void btf_bswap_type_base(struct btf_type *t)
333 {
334 t->name_off = bswap_32(t->name_off);
335 t->info = bswap_32(t->info);
336 t->type = bswap_32(t->type);
337 }
338
btf_bswap_type_rest(struct btf_type * t)339 static int btf_bswap_type_rest(struct btf_type *t)
340 {
341 struct btf_var_secinfo *v;
342 struct btf_enum64 *e64;
343 struct btf_member *m;
344 struct btf_array *a;
345 struct btf_param *p;
346 struct btf_enum *e;
347 __u16 vlen = btf_vlen(t);
348 int i;
349
350 switch (btf_kind(t)) {
351 case BTF_KIND_FWD:
352 case BTF_KIND_CONST:
353 case BTF_KIND_VOLATILE:
354 case BTF_KIND_RESTRICT:
355 case BTF_KIND_PTR:
356 case BTF_KIND_TYPEDEF:
357 case BTF_KIND_FUNC:
358 case BTF_KIND_FLOAT:
359 case BTF_KIND_TYPE_TAG:
360 return 0;
361 case BTF_KIND_INT:
362 *(__u32 *)(t + 1) = bswap_32(*(__u32 *)(t + 1));
363 return 0;
364 case BTF_KIND_ENUM:
365 for (i = 0, e = btf_enum(t); i < vlen; i++, e++) {
366 e->name_off = bswap_32(e->name_off);
367 e->val = bswap_32(e->val);
368 }
369 return 0;
370 case BTF_KIND_ENUM64:
371 for (i = 0, e64 = btf_enum64(t); i < vlen; i++, e64++) {
372 e64->name_off = bswap_32(e64->name_off);
373 e64->val_lo32 = bswap_32(e64->val_lo32);
374 e64->val_hi32 = bswap_32(e64->val_hi32);
375 }
376 return 0;
377 case BTF_KIND_ARRAY:
378 a = btf_array(t);
379 a->type = bswap_32(a->type);
380 a->index_type = bswap_32(a->index_type);
381 a->nelems = bswap_32(a->nelems);
382 return 0;
383 case BTF_KIND_STRUCT:
384 case BTF_KIND_UNION:
385 for (i = 0, m = btf_members(t); i < vlen; i++, m++) {
386 m->name_off = bswap_32(m->name_off);
387 m->type = bswap_32(m->type);
388 m->offset = bswap_32(m->offset);
389 }
390 return 0;
391 case BTF_KIND_FUNC_PROTO:
392 for (i = 0, p = btf_params(t); i < vlen; i++, p++) {
393 p->name_off = bswap_32(p->name_off);
394 p->type = bswap_32(p->type);
395 }
396 return 0;
397 case BTF_KIND_VAR:
398 btf_var(t)->linkage = bswap_32(btf_var(t)->linkage);
399 return 0;
400 case BTF_KIND_DATASEC:
401 for (i = 0, v = btf_var_secinfos(t); i < vlen; i++, v++) {
402 v->type = bswap_32(v->type);
403 v->offset = bswap_32(v->offset);
404 v->size = bswap_32(v->size);
405 }
406 return 0;
407 case BTF_KIND_DECL_TAG:
408 btf_decl_tag(t)->component_idx = bswap_32(btf_decl_tag(t)->component_idx);
409 return 0;
410 default:
411 pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
412 return -EINVAL;
413 }
414 }
415
btf_parse_type_sec(struct btf * btf)416 static int btf_parse_type_sec(struct btf *btf)
417 {
418 struct btf_header *hdr = btf->hdr;
419 void *next_type = btf->types_data;
420 void *end_type = next_type + hdr->type_len;
421 int err, type_size;
422
423 while (next_type + sizeof(struct btf_type) <= end_type) {
424 if (btf->swapped_endian)
425 btf_bswap_type_base(next_type);
426
427 type_size = btf_type_size(next_type);
428 if (type_size < 0)
429 return type_size;
430 if (next_type + type_size > end_type) {
431 pr_warn("BTF type [%d] is malformed\n", btf->start_id + btf->nr_types);
432 return -EINVAL;
433 }
434
435 if (btf->swapped_endian && btf_bswap_type_rest(next_type))
436 return -EINVAL;
437
438 err = btf_add_type_idx_entry(btf, next_type - btf->types_data);
439 if (err)
440 return err;
441
442 next_type += type_size;
443 btf->nr_types++;
444 }
445
446 if (next_type != end_type) {
447 pr_warn("BTF types data is malformed\n");
448 return -EINVAL;
449 }
450
451 return 0;
452 }
453
btf_validate_str(const struct btf * btf,__u32 str_off,const char * what,__u32 type_id)454 static int btf_validate_str(const struct btf *btf, __u32 str_off, const char *what, __u32 type_id)
455 {
456 const char *s;
457
458 s = btf__str_by_offset(btf, str_off);
459 if (!s) {
460 pr_warn("btf: type [%u]: invalid %s (string offset %u)\n", type_id, what, str_off);
461 return -EINVAL;
462 }
463
464 return 0;
465 }
466
btf_validate_id(const struct btf * btf,__u32 id,__u32 ctx_id)467 static int btf_validate_id(const struct btf *btf, __u32 id, __u32 ctx_id)
468 {
469 const struct btf_type *t;
470
471 t = btf__type_by_id(btf, id);
472 if (!t) {
473 pr_warn("btf: type [%u]: invalid referenced type ID %u\n", ctx_id, id);
474 return -EINVAL;
475 }
476
477 return 0;
478 }
479
btf_validate_type(const struct btf * btf,const struct btf_type * t,__u32 id)480 static int btf_validate_type(const struct btf *btf, const struct btf_type *t, __u32 id)
481 {
482 __u32 kind = btf_kind(t);
483 int err, i, n;
484
485 err = btf_validate_str(btf, t->name_off, "type name", id);
486 if (err)
487 return err;
488
489 switch (kind) {
490 case BTF_KIND_UNKN:
491 case BTF_KIND_INT:
492 case BTF_KIND_FWD:
493 case BTF_KIND_FLOAT:
494 break;
495 case BTF_KIND_PTR:
496 case BTF_KIND_TYPEDEF:
497 case BTF_KIND_VOLATILE:
498 case BTF_KIND_CONST:
499 case BTF_KIND_RESTRICT:
500 case BTF_KIND_VAR:
501 case BTF_KIND_DECL_TAG:
502 case BTF_KIND_TYPE_TAG:
503 err = btf_validate_id(btf, t->type, id);
504 if (err)
505 return err;
506 break;
507 case BTF_KIND_ARRAY: {
508 const struct btf_array *a = btf_array(t);
509
510 err = btf_validate_id(btf, a->type, id);
511 err = err ?: btf_validate_id(btf, a->index_type, id);
512 if (err)
513 return err;
514 break;
515 }
516 case BTF_KIND_STRUCT:
517 case BTF_KIND_UNION: {
518 const struct btf_member *m = btf_members(t);
519
520 n = btf_vlen(t);
521 for (i = 0; i < n; i++, m++) {
522 err = btf_validate_str(btf, m->name_off, "field name", id);
523 err = err ?: btf_validate_id(btf, m->type, id);
524 if (err)
525 return err;
526 }
527 break;
528 }
529 case BTF_KIND_ENUM: {
530 const struct btf_enum *m = btf_enum(t);
531
532 n = btf_vlen(t);
533 for (i = 0; i < n; i++, m++) {
534 err = btf_validate_str(btf, m->name_off, "enum name", id);
535 if (err)
536 return err;
537 }
538 break;
539 }
540 case BTF_KIND_ENUM64: {
541 const struct btf_enum64 *m = btf_enum64(t);
542
543 n = btf_vlen(t);
544 for (i = 0; i < n; i++, m++) {
545 err = btf_validate_str(btf, m->name_off, "enum name", id);
546 if (err)
547 return err;
548 }
549 break;
550 }
551 case BTF_KIND_FUNC: {
552 const struct btf_type *ft;
553
554 err = btf_validate_id(btf, t->type, id);
555 if (err)
556 return err;
557 ft = btf__type_by_id(btf, t->type);
558 if (btf_kind(ft) != BTF_KIND_FUNC_PROTO) {
559 pr_warn("btf: type [%u]: referenced type [%u] is not FUNC_PROTO\n", id, t->type);
560 return -EINVAL;
561 }
562 break;
563 }
564 case BTF_KIND_FUNC_PROTO: {
565 const struct btf_param *m = btf_params(t);
566
567 n = btf_vlen(t);
568 for (i = 0; i < n; i++, m++) {
569 err = btf_validate_str(btf, m->name_off, "param name", id);
570 err = err ?: btf_validate_id(btf, m->type, id);
571 if (err)
572 return err;
573 }
574 break;
575 }
576 case BTF_KIND_DATASEC: {
577 const struct btf_var_secinfo *m = btf_var_secinfos(t);
578
579 n = btf_vlen(t);
580 for (i = 0; i < n; i++, m++) {
581 err = btf_validate_id(btf, m->type, id);
582 if (err)
583 return err;
584 }
585 break;
586 }
587 default:
588 pr_warn("btf: type [%u]: unrecognized kind %u\n", id, kind);
589 return -EINVAL;
590 }
591 return 0;
592 }
593
594 /* Validate basic sanity of BTF. It's intentionally less thorough than
595 * kernel's validation and validates only properties of BTF that libbpf relies
596 * on to be correct (e.g., valid type IDs, valid string offsets, etc)
597 */
btf_sanity_check(const struct btf * btf)598 static int btf_sanity_check(const struct btf *btf)
599 {
600 const struct btf_type *t;
601 __u32 i, n = btf__type_cnt(btf);
602 int err;
603
604 for (i = btf->start_id; i < n; i++) {
605 t = btf_type_by_id(btf, i);
606 err = btf_validate_type(btf, t, i);
607 if (err)
608 return err;
609 }
610 return 0;
611 }
612
btf__type_cnt(const struct btf * btf)613 __u32 btf__type_cnt(const struct btf *btf)
614 {
615 return btf->start_id + btf->nr_types;
616 }
617
btf__base_btf(const struct btf * btf)618 const struct btf *btf__base_btf(const struct btf *btf)
619 {
620 return btf->base_btf;
621 }
622
623 /* internal helper returning non-const pointer to a type */
btf_type_by_id(const struct btf * btf,__u32 type_id)624 struct btf_type *btf_type_by_id(const struct btf *btf, __u32 type_id)
625 {
626 if (type_id == 0)
627 return &btf_void;
628 if (type_id < btf->start_id)
629 return btf_type_by_id(btf->base_btf, type_id);
630 return btf->types_data + btf->type_offs[type_id - btf->start_id];
631 }
632
btf__type_by_id(const struct btf * btf,__u32 type_id)633 const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
634 {
635 if (type_id >= btf->start_id + btf->nr_types)
636 return errno = EINVAL, NULL;
637 return btf_type_by_id((struct btf *)btf, type_id);
638 }
639
determine_ptr_size(const struct btf * btf)640 static int determine_ptr_size(const struct btf *btf)
641 {
642 static const char * const long_aliases[] = {
643 "long",
644 "long int",
645 "int long",
646 "unsigned long",
647 "long unsigned",
648 "unsigned long int",
649 "unsigned int long",
650 "long unsigned int",
651 "long int unsigned",
652 "int unsigned long",
653 "int long unsigned",
654 };
655 const struct btf_type *t;
656 const char *name;
657 int i, j, n;
658
659 if (btf->base_btf && btf->base_btf->ptr_sz > 0)
660 return btf->base_btf->ptr_sz;
661
662 n = btf__type_cnt(btf);
663 for (i = 1; i < n; i++) {
664 t = btf__type_by_id(btf, i);
665 if (!btf_is_int(t))
666 continue;
667
668 if (t->size != 4 && t->size != 8)
669 continue;
670
671 name = btf__name_by_offset(btf, t->name_off);
672 if (!name)
673 continue;
674
675 for (j = 0; j < ARRAY_SIZE(long_aliases); j++) {
676 if (strcmp(name, long_aliases[j]) == 0)
677 return t->size;
678 }
679 }
680
681 return -1;
682 }
683
btf_ptr_sz(const struct btf * btf)684 static size_t btf_ptr_sz(const struct btf *btf)
685 {
686 if (!btf->ptr_sz)
687 ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
688 return btf->ptr_sz < 0 ? sizeof(void *) : btf->ptr_sz;
689 }
690
691 /* Return pointer size this BTF instance assumes. The size is heuristically
692 * determined by looking for 'long' or 'unsigned long' integer type and
693 * recording its size in bytes. If BTF type information doesn't have any such
694 * type, this function returns 0. In the latter case, native architecture's
695 * pointer size is assumed, so will be either 4 or 8, depending on
696 * architecture that libbpf was compiled for. It's possible to override
697 * guessed value by using btf__set_pointer_size() API.
698 */
btf__pointer_size(const struct btf * btf)699 size_t btf__pointer_size(const struct btf *btf)
700 {
701 if (!btf->ptr_sz)
702 ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
703
704 if (btf->ptr_sz < 0)
705 /* not enough BTF type info to guess */
706 return 0;
707
708 return btf->ptr_sz;
709 }
710
711 /* Override or set pointer size in bytes. Only values of 4 and 8 are
712 * supported.
713 */
btf__set_pointer_size(struct btf * btf,size_t ptr_sz)714 int btf__set_pointer_size(struct btf *btf, size_t ptr_sz)
715 {
716 if (ptr_sz != 4 && ptr_sz != 8)
717 return libbpf_err(-EINVAL);
718 btf->ptr_sz = ptr_sz;
719 return 0;
720 }
721
is_host_big_endian(void)722 static bool is_host_big_endian(void)
723 {
724 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
725 return false;
726 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
727 return true;
728 #else
729 # error "Unrecognized __BYTE_ORDER__"
730 #endif
731 }
732
btf__endianness(const struct btf * btf)733 enum btf_endianness btf__endianness(const struct btf *btf)
734 {
735 if (is_host_big_endian())
736 return btf->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
737 else
738 return btf->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
739 }
740
btf__set_endianness(struct btf * btf,enum btf_endianness endian)741 int btf__set_endianness(struct btf *btf, enum btf_endianness endian)
742 {
743 if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
744 return libbpf_err(-EINVAL);
745
746 btf->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
747 if (!btf->swapped_endian) {
748 free(btf->raw_data_swapped);
749 btf->raw_data_swapped = NULL;
750 }
751 return 0;
752 }
753
btf_type_is_void(const struct btf_type * t)754 static bool btf_type_is_void(const struct btf_type *t)
755 {
756 return t == &btf_void || btf_is_fwd(t);
757 }
758
btf_type_is_void_or_null(const struct btf_type * t)759 static bool btf_type_is_void_or_null(const struct btf_type *t)
760 {
761 return !t || btf_type_is_void(t);
762 }
763
764 #define MAX_RESOLVE_DEPTH 32
765
btf__resolve_size(const struct btf * btf,__u32 type_id)766 __s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
767 {
768 const struct btf_array *array;
769 const struct btf_type *t;
770 __u32 nelems = 1;
771 __s64 size = -1;
772 int i;
773
774 t = btf__type_by_id(btf, type_id);
775 for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t); i++) {
776 switch (btf_kind(t)) {
777 case BTF_KIND_INT:
778 case BTF_KIND_STRUCT:
779 case BTF_KIND_UNION:
780 case BTF_KIND_ENUM:
781 case BTF_KIND_ENUM64:
782 case BTF_KIND_DATASEC:
783 case BTF_KIND_FLOAT:
784 size = t->size;
785 goto done;
786 case BTF_KIND_PTR:
787 size = btf_ptr_sz(btf);
788 goto done;
789 case BTF_KIND_TYPEDEF:
790 case BTF_KIND_VOLATILE:
791 case BTF_KIND_CONST:
792 case BTF_KIND_RESTRICT:
793 case BTF_KIND_VAR:
794 case BTF_KIND_DECL_TAG:
795 case BTF_KIND_TYPE_TAG:
796 type_id = t->type;
797 break;
798 case BTF_KIND_ARRAY:
799 array = btf_array(t);
800 if (nelems && array->nelems > UINT32_MAX / nelems)
801 return libbpf_err(-E2BIG);
802 nelems *= array->nelems;
803 type_id = array->type;
804 break;
805 default:
806 return libbpf_err(-EINVAL);
807 }
808
809 t = btf__type_by_id(btf, type_id);
810 }
811
812 done:
813 if (size < 0)
814 return libbpf_err(-EINVAL);
815 if (nelems && size > UINT32_MAX / nelems)
816 return libbpf_err(-E2BIG);
817
818 return nelems * size;
819 }
820
btf__align_of(const struct btf * btf,__u32 id)821 int btf__align_of(const struct btf *btf, __u32 id)
822 {
823 const struct btf_type *t = btf__type_by_id(btf, id);
824 __u16 kind = btf_kind(t);
825
826 switch (kind) {
827 case BTF_KIND_INT:
828 case BTF_KIND_ENUM:
829 case BTF_KIND_ENUM64:
830 case BTF_KIND_FLOAT:
831 return min(btf_ptr_sz(btf), (size_t)t->size);
832 case BTF_KIND_PTR:
833 return btf_ptr_sz(btf);
834 case BTF_KIND_TYPEDEF:
835 case BTF_KIND_VOLATILE:
836 case BTF_KIND_CONST:
837 case BTF_KIND_RESTRICT:
838 case BTF_KIND_TYPE_TAG:
839 return btf__align_of(btf, t->type);
840 case BTF_KIND_ARRAY:
841 return btf__align_of(btf, btf_array(t)->type);
842 case BTF_KIND_STRUCT:
843 case BTF_KIND_UNION: {
844 const struct btf_member *m = btf_members(t);
845 __u16 vlen = btf_vlen(t);
846 int i, max_align = 1, align;
847
848 for (i = 0; i < vlen; i++, m++) {
849 align = btf__align_of(btf, m->type);
850 if (align <= 0)
851 return libbpf_err(align);
852 max_align = max(max_align, align);
853
854 /* if field offset isn't aligned according to field
855 * type's alignment, then struct must be packed
856 */
857 if (btf_member_bitfield_size(t, i) == 0 &&
858 (m->offset % (8 * align)) != 0)
859 return 1;
860 }
861
862 /* if struct/union size isn't a multiple of its alignment,
863 * then struct must be packed
864 */
865 if ((t->size % max_align) != 0)
866 return 1;
867
868 return max_align;
869 }
870 default:
871 pr_warn("unsupported BTF_KIND:%u\n", btf_kind(t));
872 return errno = EINVAL, 0;
873 }
874 }
875
btf__resolve_type(const struct btf * btf,__u32 type_id)876 int btf__resolve_type(const struct btf *btf, __u32 type_id)
877 {
878 const struct btf_type *t;
879 int depth = 0;
880
881 t = btf__type_by_id(btf, type_id);
882 while (depth < MAX_RESOLVE_DEPTH &&
883 !btf_type_is_void_or_null(t) &&
884 (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
885 type_id = t->type;
886 t = btf__type_by_id(btf, type_id);
887 depth++;
888 }
889
890 if (depth == MAX_RESOLVE_DEPTH || btf_type_is_void_or_null(t))
891 return libbpf_err(-EINVAL);
892
893 return type_id;
894 }
895
btf__find_by_name(const struct btf * btf,const char * type_name)896 __s32 btf__find_by_name(const struct btf *btf, const char *type_name)
897 {
898 __u32 i, nr_types = btf__type_cnt(btf);
899
900 if (!strcmp(type_name, "void"))
901 return 0;
902
903 for (i = 1; i < nr_types; i++) {
904 const struct btf_type *t = btf__type_by_id(btf, i);
905 const char *name = btf__name_by_offset(btf, t->name_off);
906
907 if (name && !strcmp(type_name, name))
908 return i;
909 }
910
911 return libbpf_err(-ENOENT);
912 }
913
btf_find_by_name_kind(const struct btf * btf,int start_id,const char * type_name,__u32 kind)914 static __s32 btf_find_by_name_kind(const struct btf *btf, int start_id,
915 const char *type_name, __u32 kind)
916 {
917 __u32 i, nr_types = btf__type_cnt(btf);
918
919 if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
920 return 0;
921
922 for (i = start_id; i < nr_types; i++) {
923 const struct btf_type *t = btf__type_by_id(btf, i);
924 const char *name;
925
926 if (btf_kind(t) != kind)
927 continue;
928 name = btf__name_by_offset(btf, t->name_off);
929 if (name && !strcmp(type_name, name))
930 return i;
931 }
932
933 return libbpf_err(-ENOENT);
934 }
935
btf__find_by_name_kind_own(const struct btf * btf,const char * type_name,__u32 kind)936 __s32 btf__find_by_name_kind_own(const struct btf *btf, const char *type_name,
937 __u32 kind)
938 {
939 return btf_find_by_name_kind(btf, btf->start_id, type_name, kind);
940 }
941
btf__find_by_name_kind(const struct btf * btf,const char * type_name,__u32 kind)942 __s32 btf__find_by_name_kind(const struct btf *btf, const char *type_name,
943 __u32 kind)
944 {
945 return btf_find_by_name_kind(btf, 1, type_name, kind);
946 }
947
btf_is_modifiable(const struct btf * btf)948 static bool btf_is_modifiable(const struct btf *btf)
949 {
950 return (void *)btf->hdr != btf->raw_data;
951 }
952
btf__free(struct btf * btf)953 void btf__free(struct btf *btf)
954 {
955 if (IS_ERR_OR_NULL(btf))
956 return;
957
958 if (btf->fd >= 0)
959 close(btf->fd);
960
961 if (btf_is_modifiable(btf)) {
962 /* if BTF was modified after loading, it will have a split
963 * in-memory representation for header, types, and strings
964 * sections, so we need to free all of them individually. It
965 * might still have a cached contiguous raw data present,
966 * which will be unconditionally freed below.
967 */
968 free(btf->hdr);
969 free(btf->types_data);
970 strset__free(btf->strs_set);
971 }
972 free(btf->raw_data);
973 free(btf->raw_data_swapped);
974 free(btf->type_offs);
975 if (btf->owns_base)
976 btf__free(btf->base_btf);
977 free(btf);
978 }
979
btf_new_empty(struct btf * base_btf)980 static struct btf *btf_new_empty(struct btf *base_btf)
981 {
982 struct btf *btf;
983
984 btf = calloc(1, sizeof(*btf));
985 if (!btf)
986 return ERR_PTR(-ENOMEM);
987
988 btf->nr_types = 0;
989 btf->start_id = 1;
990 btf->start_str_off = 0;
991 btf->fd = -1;
992 btf->ptr_sz = sizeof(void *);
993 btf->swapped_endian = false;
994
995 if (base_btf) {
996 btf->base_btf = base_btf;
997 btf->start_id = btf__type_cnt(base_btf);
998 btf->start_str_off = base_btf->hdr->str_len + base_btf->start_str_off;
999 btf->swapped_endian = base_btf->swapped_endian;
1000 }
1001
1002 /* +1 for empty string at offset 0 */
1003 btf->raw_size = sizeof(struct btf_header) + (base_btf ? 0 : 1);
1004 btf->raw_data = calloc(1, btf->raw_size);
1005 if (!btf->raw_data) {
1006 free(btf);
1007 return ERR_PTR(-ENOMEM);
1008 }
1009
1010 btf->hdr = btf->raw_data;
1011 btf->hdr->hdr_len = sizeof(struct btf_header);
1012 btf->hdr->magic = BTF_MAGIC;
1013 btf->hdr->version = BTF_VERSION;
1014
1015 btf->types_data = btf->raw_data + btf->hdr->hdr_len;
1016 btf->strs_data = btf->raw_data + btf->hdr->hdr_len;
1017 btf->hdr->str_len = base_btf ? 0 : 1; /* empty string at offset 0 */
1018
1019 return btf;
1020 }
1021
btf__new_empty(void)1022 struct btf *btf__new_empty(void)
1023 {
1024 return libbpf_ptr(btf_new_empty(NULL));
1025 }
1026
btf__new_empty_split(struct btf * base_btf)1027 struct btf *btf__new_empty_split(struct btf *base_btf)
1028 {
1029 return libbpf_ptr(btf_new_empty(base_btf));
1030 }
1031
btf_new(const void * data,__u32 size,struct btf * base_btf)1032 static struct btf *btf_new(const void *data, __u32 size, struct btf *base_btf)
1033 {
1034 struct btf *btf;
1035 int err;
1036
1037 btf = calloc(1, sizeof(struct btf));
1038 if (!btf)
1039 return ERR_PTR(-ENOMEM);
1040
1041 btf->nr_types = 0;
1042 btf->start_id = 1;
1043 btf->start_str_off = 0;
1044 btf->fd = -1;
1045
1046 if (base_btf) {
1047 btf->base_btf = base_btf;
1048 btf->start_id = btf__type_cnt(base_btf);
1049 btf->start_str_off = base_btf->hdr->str_len;
1050 }
1051
1052 btf->raw_data = malloc(size);
1053 if (!btf->raw_data) {
1054 err = -ENOMEM;
1055 goto done;
1056 }
1057 memcpy(btf->raw_data, data, size);
1058 btf->raw_size = size;
1059
1060 btf->hdr = btf->raw_data;
1061 err = btf_parse_hdr(btf);
1062 if (err)
1063 goto done;
1064
1065 btf->strs_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->str_off;
1066 btf->types_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->type_off;
1067
1068 err = btf_parse_str_sec(btf);
1069 err = err ?: btf_parse_type_sec(btf);
1070 err = err ?: btf_sanity_check(btf);
1071 if (err)
1072 goto done;
1073
1074 done:
1075 if (err) {
1076 btf__free(btf);
1077 return ERR_PTR(err);
1078 }
1079
1080 return btf;
1081 }
1082
btf__new(const void * data,__u32 size)1083 struct btf *btf__new(const void *data, __u32 size)
1084 {
1085 return libbpf_ptr(btf_new(data, size, NULL));
1086 }
1087
btf__new_split(const void * data,__u32 size,struct btf * base_btf)1088 struct btf *btf__new_split(const void *data, __u32 size, struct btf *base_btf)
1089 {
1090 return libbpf_ptr(btf_new(data, size, base_btf));
1091 }
1092
1093 struct btf_elf_secs {
1094 Elf_Data *btf_data;
1095 Elf_Data *btf_ext_data;
1096 Elf_Data *btf_base_data;
1097 };
1098
btf_find_elf_sections(Elf * elf,const char * path,struct btf_elf_secs * secs)1099 static int btf_find_elf_sections(Elf *elf, const char *path, struct btf_elf_secs *secs)
1100 {
1101 Elf_Scn *scn = NULL;
1102 Elf_Data *data;
1103 GElf_Ehdr ehdr;
1104 size_t shstrndx;
1105 int idx = 0;
1106
1107 if (!gelf_getehdr(elf, &ehdr)) {
1108 pr_warn("failed to get EHDR from %s\n", path);
1109 goto err;
1110 }
1111
1112 if (elf_getshdrstrndx(elf, &shstrndx)) {
1113 pr_warn("failed to get section names section index for %s\n",
1114 path);
1115 goto err;
1116 }
1117
1118 if (!elf_rawdata(elf_getscn(elf, shstrndx), NULL)) {
1119 pr_warn("failed to get e_shstrndx from %s\n", path);
1120 goto err;
1121 }
1122
1123 while ((scn = elf_nextscn(elf, scn)) != NULL) {
1124 Elf_Data **field;
1125 GElf_Shdr sh;
1126 char *name;
1127
1128 idx++;
1129 if (gelf_getshdr(scn, &sh) != &sh) {
1130 pr_warn("failed to get section(%d) header from %s\n",
1131 idx, path);
1132 goto err;
1133 }
1134 name = elf_strptr(elf, shstrndx, sh.sh_name);
1135 if (!name) {
1136 pr_warn("failed to get section(%d) name from %s\n",
1137 idx, path);
1138 goto err;
1139 }
1140
1141 if (strcmp(name, BTF_ELF_SEC) == 0)
1142 field = &secs->btf_data;
1143 else if (strcmp(name, BTF_EXT_ELF_SEC) == 0)
1144 field = &secs->btf_ext_data;
1145 else if (strcmp(name, BTF_BASE_ELF_SEC) == 0)
1146 field = &secs->btf_base_data;
1147 else
1148 continue;
1149
1150 data = elf_getdata(scn, 0);
1151 if (!data) {
1152 pr_warn("failed to get section(%d, %s) data from %s\n",
1153 idx, name, path);
1154 goto err;
1155 }
1156 *field = data;
1157 }
1158
1159 return 0;
1160
1161 err:
1162 return -LIBBPF_ERRNO__FORMAT;
1163 }
1164
btf_parse_elf(const char * path,struct btf * base_btf,struct btf_ext ** btf_ext)1165 static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,
1166 struct btf_ext **btf_ext)
1167 {
1168 struct btf_elf_secs secs = {};
1169 struct btf *dist_base_btf = NULL;
1170 struct btf *btf = NULL;
1171 int err = 0, fd = -1;
1172 Elf *elf = NULL;
1173
1174 if (elf_version(EV_CURRENT) == EV_NONE) {
1175 pr_warn("failed to init libelf for %s\n", path);
1176 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
1177 }
1178
1179 fd = open(path, O_RDONLY | O_CLOEXEC);
1180 if (fd < 0) {
1181 err = -errno;
1182 pr_warn("failed to open %s: %s\n", path, strerror(errno));
1183 return ERR_PTR(err);
1184 }
1185
1186 elf = elf_begin(fd, ELF_C_READ, NULL);
1187 if (!elf) {
1188 err = -LIBBPF_ERRNO__FORMAT;
1189 pr_warn("failed to open %s as ELF file\n", path);
1190 goto done;
1191 }
1192
1193 err = btf_find_elf_sections(elf, path, &secs);
1194 if (err)
1195 goto done;
1196
1197 if (!secs.btf_data) {
1198 pr_warn("failed to find '%s' ELF section in %s\n", BTF_ELF_SEC, path);
1199 err = -ENODATA;
1200 goto done;
1201 }
1202
1203 if (secs.btf_base_data) {
1204 dist_base_btf = btf_new(secs.btf_base_data->d_buf, secs.btf_base_data->d_size,
1205 NULL);
1206 if (IS_ERR(dist_base_btf)) {
1207 err = PTR_ERR(dist_base_btf);
1208 dist_base_btf = NULL;
1209 goto done;
1210 }
1211 }
1212
1213 btf = btf_new(secs.btf_data->d_buf, secs.btf_data->d_size,
1214 dist_base_btf ?: base_btf);
1215 if (IS_ERR(btf)) {
1216 err = PTR_ERR(btf);
1217 goto done;
1218 }
1219 if (dist_base_btf && base_btf) {
1220 err = btf__relocate(btf, base_btf);
1221 if (err)
1222 goto done;
1223 btf__free(dist_base_btf);
1224 dist_base_btf = NULL;
1225 }
1226
1227 if (dist_base_btf)
1228 btf->owns_base = true;
1229
1230 switch (gelf_getclass(elf)) {
1231 case ELFCLASS32:
1232 btf__set_pointer_size(btf, 4);
1233 break;
1234 case ELFCLASS64:
1235 btf__set_pointer_size(btf, 8);
1236 break;
1237 default:
1238 pr_warn("failed to get ELF class (bitness) for %s\n", path);
1239 break;
1240 }
1241
1242 if (btf_ext && secs.btf_ext_data) {
1243 *btf_ext = btf_ext__new(secs.btf_ext_data->d_buf, secs.btf_ext_data->d_size);
1244 if (IS_ERR(*btf_ext)) {
1245 err = PTR_ERR(*btf_ext);
1246 goto done;
1247 }
1248 } else if (btf_ext) {
1249 *btf_ext = NULL;
1250 }
1251 done:
1252 if (elf)
1253 elf_end(elf);
1254 close(fd);
1255
1256 if (!err)
1257 return btf;
1258
1259 if (btf_ext)
1260 btf_ext__free(*btf_ext);
1261 btf__free(dist_base_btf);
1262 btf__free(btf);
1263
1264 return ERR_PTR(err);
1265 }
1266
btf__parse_elf(const char * path,struct btf_ext ** btf_ext)1267 struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
1268 {
1269 return libbpf_ptr(btf_parse_elf(path, NULL, btf_ext));
1270 }
1271
btf__parse_elf_split(const char * path,struct btf * base_btf)1272 struct btf *btf__parse_elf_split(const char *path, struct btf *base_btf)
1273 {
1274 return libbpf_ptr(btf_parse_elf(path, base_btf, NULL));
1275 }
1276
btf_parse_raw(const char * path,struct btf * base_btf)1277 static struct btf *btf_parse_raw(const char *path, struct btf *base_btf)
1278 {
1279 struct btf *btf = NULL;
1280 void *data = NULL;
1281 FILE *f = NULL;
1282 __u16 magic;
1283 int err = 0;
1284 long sz;
1285
1286 f = fopen(path, "rbe");
1287 if (!f) {
1288 err = -errno;
1289 goto err_out;
1290 }
1291
1292 /* check BTF magic */
1293 if (fread(&magic, 1, sizeof(magic), f) < sizeof(magic)) {
1294 err = -EIO;
1295 goto err_out;
1296 }
1297 if (magic != BTF_MAGIC && magic != bswap_16(BTF_MAGIC)) {
1298 /* definitely not a raw BTF */
1299 err = -EPROTO;
1300 goto err_out;
1301 }
1302
1303 /* get file size */
1304 if (fseek(f, 0, SEEK_END)) {
1305 err = -errno;
1306 goto err_out;
1307 }
1308 sz = ftell(f);
1309 if (sz < 0) {
1310 err = -errno;
1311 goto err_out;
1312 }
1313 /* rewind to the start */
1314 if (fseek(f, 0, SEEK_SET)) {
1315 err = -errno;
1316 goto err_out;
1317 }
1318
1319 /* pre-alloc memory and read all of BTF data */
1320 data = malloc(sz);
1321 if (!data) {
1322 err = -ENOMEM;
1323 goto err_out;
1324 }
1325 if (fread(data, 1, sz, f) < sz) {
1326 err = -EIO;
1327 goto err_out;
1328 }
1329
1330 /* finally parse BTF data */
1331 btf = btf_new(data, sz, base_btf);
1332
1333 err_out:
1334 free(data);
1335 if (f)
1336 fclose(f);
1337 return err ? ERR_PTR(err) : btf;
1338 }
1339
btf__parse_raw(const char * path)1340 struct btf *btf__parse_raw(const char *path)
1341 {
1342 return libbpf_ptr(btf_parse_raw(path, NULL));
1343 }
1344
btf__parse_raw_split(const char * path,struct btf * base_btf)1345 struct btf *btf__parse_raw_split(const char *path, struct btf *base_btf)
1346 {
1347 return libbpf_ptr(btf_parse_raw(path, base_btf));
1348 }
1349
btf_parse(const char * path,struct btf * base_btf,struct btf_ext ** btf_ext)1350 static struct btf *btf_parse(const char *path, struct btf *base_btf, struct btf_ext **btf_ext)
1351 {
1352 struct btf *btf;
1353 int err;
1354
1355 if (btf_ext)
1356 *btf_ext = NULL;
1357
1358 btf = btf_parse_raw(path, base_btf);
1359 err = libbpf_get_error(btf);
1360 if (!err)
1361 return btf;
1362 if (err != -EPROTO)
1363 return ERR_PTR(err);
1364 return btf_parse_elf(path, base_btf, btf_ext);
1365 }
1366
btf__parse(const char * path,struct btf_ext ** btf_ext)1367 struct btf *btf__parse(const char *path, struct btf_ext **btf_ext)
1368 {
1369 return libbpf_ptr(btf_parse(path, NULL, btf_ext));
1370 }
1371
btf__parse_split(const char * path,struct btf * base_btf)1372 struct btf *btf__parse_split(const char *path, struct btf *base_btf)
1373 {
1374 return libbpf_ptr(btf_parse(path, base_btf, NULL));
1375 }
1376
1377 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian);
1378
btf_load_into_kernel(struct btf * btf,char * log_buf,size_t log_sz,__u32 log_level,int token_fd)1379 int btf_load_into_kernel(struct btf *btf,
1380 char *log_buf, size_t log_sz, __u32 log_level,
1381 int token_fd)
1382 {
1383 LIBBPF_OPTS(bpf_btf_load_opts, opts);
1384 __u32 buf_sz = 0, raw_size;
1385 char *buf = NULL, *tmp;
1386 void *raw_data;
1387 int err = 0;
1388
1389 if (btf->fd >= 0)
1390 return libbpf_err(-EEXIST);
1391 if (log_sz && !log_buf)
1392 return libbpf_err(-EINVAL);
1393
1394 /* cache native raw data representation */
1395 raw_data = btf_get_raw_data(btf, &raw_size, false);
1396 if (!raw_data) {
1397 err = -ENOMEM;
1398 goto done;
1399 }
1400 btf->raw_size = raw_size;
1401 btf->raw_data = raw_data;
1402
1403 retry_load:
1404 /* if log_level is 0, we won't provide log_buf/log_size to the kernel,
1405 * initially. Only if BTF loading fails, we bump log_level to 1 and
1406 * retry, using either auto-allocated or custom log_buf. This way
1407 * non-NULL custom log_buf provides a buffer just in case, but hopes
1408 * for successful load and no need for log_buf.
1409 */
1410 if (log_level) {
1411 /* if caller didn't provide custom log_buf, we'll keep
1412 * allocating our own progressively bigger buffers for BTF
1413 * verification log
1414 */
1415 if (!log_buf) {
1416 buf_sz = max((__u32)BPF_LOG_BUF_SIZE, buf_sz * 2);
1417 tmp = realloc(buf, buf_sz);
1418 if (!tmp) {
1419 err = -ENOMEM;
1420 goto done;
1421 }
1422 buf = tmp;
1423 buf[0] = '\0';
1424 }
1425
1426 opts.log_buf = log_buf ? log_buf : buf;
1427 opts.log_size = log_buf ? log_sz : buf_sz;
1428 opts.log_level = log_level;
1429 }
1430
1431 opts.token_fd = token_fd;
1432 if (token_fd)
1433 opts.btf_flags |= BPF_F_TOKEN_FD;
1434
1435 btf->fd = bpf_btf_load(raw_data, raw_size, &opts);
1436 if (btf->fd < 0) {
1437 /* time to turn on verbose mode and try again */
1438 if (log_level == 0) {
1439 log_level = 1;
1440 goto retry_load;
1441 }
1442 /* only retry if caller didn't provide custom log_buf, but
1443 * make sure we can never overflow buf_sz
1444 */
1445 if (!log_buf && errno == ENOSPC && buf_sz <= UINT_MAX / 2)
1446 goto retry_load;
1447
1448 err = -errno;
1449 pr_warn("BTF loading error: %d\n", err);
1450 /* don't print out contents of custom log_buf */
1451 if (!log_buf && buf[0])
1452 pr_warn("-- BEGIN BTF LOAD LOG ---\n%s\n-- END BTF LOAD LOG --\n", buf);
1453 }
1454
1455 done:
1456 free(buf);
1457 return libbpf_err(err);
1458 }
1459
btf__load_into_kernel(struct btf * btf)1460 int btf__load_into_kernel(struct btf *btf)
1461 {
1462 return btf_load_into_kernel(btf, NULL, 0, 0, 0);
1463 }
1464
btf__fd(const struct btf * btf)1465 int btf__fd(const struct btf *btf)
1466 {
1467 return btf->fd;
1468 }
1469
btf__set_fd(struct btf * btf,int fd)1470 void btf__set_fd(struct btf *btf, int fd)
1471 {
1472 btf->fd = fd;
1473 }
1474
btf_strs_data(const struct btf * btf)1475 static const void *btf_strs_data(const struct btf *btf)
1476 {
1477 return btf->strs_data ? btf->strs_data : strset__data(btf->strs_set);
1478 }
1479
btf_get_raw_data(const struct btf * btf,__u32 * size,bool swap_endian)1480 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian)
1481 {
1482 struct btf_header *hdr = btf->hdr;
1483 struct btf_type *t;
1484 void *data, *p;
1485 __u32 data_sz;
1486 int i;
1487
1488 data = swap_endian ? btf->raw_data_swapped : btf->raw_data;
1489 if (data) {
1490 *size = btf->raw_size;
1491 return data;
1492 }
1493
1494 data_sz = hdr->hdr_len + hdr->type_len + hdr->str_len;
1495 data = calloc(1, data_sz);
1496 if (!data)
1497 return NULL;
1498 p = data;
1499
1500 memcpy(p, hdr, hdr->hdr_len);
1501 if (swap_endian)
1502 btf_bswap_hdr(p);
1503 p += hdr->hdr_len;
1504
1505 memcpy(p, btf->types_data, hdr->type_len);
1506 if (swap_endian) {
1507 for (i = 0; i < btf->nr_types; i++) {
1508 t = p + btf->type_offs[i];
1509 /* btf_bswap_type_rest() relies on native t->info, so
1510 * we swap base type info after we swapped all the
1511 * additional information
1512 */
1513 if (btf_bswap_type_rest(t))
1514 goto err_out;
1515 btf_bswap_type_base(t);
1516 }
1517 }
1518 p += hdr->type_len;
1519
1520 memcpy(p, btf_strs_data(btf), hdr->str_len);
1521 p += hdr->str_len;
1522
1523 *size = data_sz;
1524 return data;
1525 err_out:
1526 free(data);
1527 return NULL;
1528 }
1529
btf__raw_data(const struct btf * btf_ro,__u32 * size)1530 const void *btf__raw_data(const struct btf *btf_ro, __u32 *size)
1531 {
1532 struct btf *btf = (struct btf *)btf_ro;
1533 __u32 data_sz;
1534 void *data;
1535
1536 data = btf_get_raw_data(btf, &data_sz, btf->swapped_endian);
1537 if (!data)
1538 return errno = ENOMEM, NULL;
1539
1540 btf->raw_size = data_sz;
1541 if (btf->swapped_endian)
1542 btf->raw_data_swapped = data;
1543 else
1544 btf->raw_data = data;
1545 *size = data_sz;
1546 return data;
1547 }
1548
1549 __attribute__((alias("btf__raw_data")))
1550 const void *btf__get_raw_data(const struct btf *btf, __u32 *size);
1551
btf__str_by_offset(const struct btf * btf,__u32 offset)1552 const char *btf__str_by_offset(const struct btf *btf, __u32 offset)
1553 {
1554 if (offset < btf->start_str_off)
1555 return btf__str_by_offset(btf->base_btf, offset);
1556 else if (offset - btf->start_str_off < btf->hdr->str_len)
1557 return btf_strs_data(btf) + (offset - btf->start_str_off);
1558 else
1559 return errno = EINVAL, NULL;
1560 }
1561
btf__name_by_offset(const struct btf * btf,__u32 offset)1562 const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
1563 {
1564 return btf__str_by_offset(btf, offset);
1565 }
1566
btf_get_from_fd(int btf_fd,struct btf * base_btf)1567 struct btf *btf_get_from_fd(int btf_fd, struct btf *base_btf)
1568 {
1569 struct bpf_btf_info btf_info;
1570 __u32 len = sizeof(btf_info);
1571 __u32 last_size;
1572 struct btf *btf;
1573 void *ptr;
1574 int err;
1575
1576 /* we won't know btf_size until we call bpf_btf_get_info_by_fd(). so
1577 * let's start with a sane default - 4KiB here - and resize it only if
1578 * bpf_btf_get_info_by_fd() needs a bigger buffer.
1579 */
1580 last_size = 4096;
1581 ptr = malloc(last_size);
1582 if (!ptr)
1583 return ERR_PTR(-ENOMEM);
1584
1585 memset(&btf_info, 0, sizeof(btf_info));
1586 btf_info.btf = ptr_to_u64(ptr);
1587 btf_info.btf_size = last_size;
1588 err = bpf_btf_get_info_by_fd(btf_fd, &btf_info, &len);
1589
1590 if (!err && btf_info.btf_size > last_size) {
1591 void *temp_ptr;
1592
1593 last_size = btf_info.btf_size;
1594 temp_ptr = realloc(ptr, last_size);
1595 if (!temp_ptr) {
1596 btf = ERR_PTR(-ENOMEM);
1597 goto exit_free;
1598 }
1599 ptr = temp_ptr;
1600
1601 len = sizeof(btf_info);
1602 memset(&btf_info, 0, sizeof(btf_info));
1603 btf_info.btf = ptr_to_u64(ptr);
1604 btf_info.btf_size = last_size;
1605
1606 err = bpf_btf_get_info_by_fd(btf_fd, &btf_info, &len);
1607 }
1608
1609 if (err || btf_info.btf_size > last_size) {
1610 btf = err ? ERR_PTR(-errno) : ERR_PTR(-E2BIG);
1611 goto exit_free;
1612 }
1613
1614 btf = btf_new(ptr, btf_info.btf_size, base_btf);
1615
1616 exit_free:
1617 free(ptr);
1618 return btf;
1619 }
1620
btf__load_from_kernel_by_id_split(__u32 id,struct btf * base_btf)1621 struct btf *btf__load_from_kernel_by_id_split(__u32 id, struct btf *base_btf)
1622 {
1623 struct btf *btf;
1624 int btf_fd;
1625
1626 btf_fd = bpf_btf_get_fd_by_id(id);
1627 if (btf_fd < 0)
1628 return libbpf_err_ptr(-errno);
1629
1630 btf = btf_get_from_fd(btf_fd, base_btf);
1631 close(btf_fd);
1632
1633 return libbpf_ptr(btf);
1634 }
1635
btf__load_from_kernel_by_id(__u32 id)1636 struct btf *btf__load_from_kernel_by_id(__u32 id)
1637 {
1638 return btf__load_from_kernel_by_id_split(id, NULL);
1639 }
1640
btf_invalidate_raw_data(struct btf * btf)1641 static void btf_invalidate_raw_data(struct btf *btf)
1642 {
1643 if (btf->raw_data) {
1644 free(btf->raw_data);
1645 btf->raw_data = NULL;
1646 }
1647 if (btf->raw_data_swapped) {
1648 free(btf->raw_data_swapped);
1649 btf->raw_data_swapped = NULL;
1650 }
1651 }
1652
1653 /* Ensure BTF is ready to be modified (by splitting into a three memory
1654 * regions for header, types, and strings). Also invalidate cached
1655 * raw_data, if any.
1656 */
btf_ensure_modifiable(struct btf * btf)1657 static int btf_ensure_modifiable(struct btf *btf)
1658 {
1659 void *hdr, *types;
1660 struct strset *set = NULL;
1661 int err = -ENOMEM;
1662
1663 if (btf_is_modifiable(btf)) {
1664 /* any BTF modification invalidates raw_data */
1665 btf_invalidate_raw_data(btf);
1666 return 0;
1667 }
1668
1669 /* split raw data into three memory regions */
1670 hdr = malloc(btf->hdr->hdr_len);
1671 types = malloc(btf->hdr->type_len);
1672 if (!hdr || !types)
1673 goto err_out;
1674
1675 memcpy(hdr, btf->hdr, btf->hdr->hdr_len);
1676 memcpy(types, btf->types_data, btf->hdr->type_len);
1677
1678 /* build lookup index for all strings */
1679 set = strset__new(BTF_MAX_STR_OFFSET, btf->strs_data, btf->hdr->str_len);
1680 if (IS_ERR(set)) {
1681 err = PTR_ERR(set);
1682 goto err_out;
1683 }
1684
1685 /* only when everything was successful, update internal state */
1686 btf->hdr = hdr;
1687 btf->types_data = types;
1688 btf->types_data_cap = btf->hdr->type_len;
1689 btf->strs_data = NULL;
1690 btf->strs_set = set;
1691 /* if BTF was created from scratch, all strings are guaranteed to be
1692 * unique and deduplicated
1693 */
1694 if (btf->hdr->str_len == 0)
1695 btf->strs_deduped = true;
1696 if (!btf->base_btf && btf->hdr->str_len == 1)
1697 btf->strs_deduped = true;
1698
1699 /* invalidate raw_data representation */
1700 btf_invalidate_raw_data(btf);
1701
1702 return 0;
1703
1704 err_out:
1705 strset__free(set);
1706 free(hdr);
1707 free(types);
1708 return err;
1709 }
1710
1711 /* Find an offset in BTF string section that corresponds to a given string *s*.
1712 * Returns:
1713 * - >0 offset into string section, if string is found;
1714 * - -ENOENT, if string is not in the string section;
1715 * - <0, on any other error.
1716 */
btf__find_str(struct btf * btf,const char * s)1717 int btf__find_str(struct btf *btf, const char *s)
1718 {
1719 int off;
1720
1721 if (btf->base_btf) {
1722 off = btf__find_str(btf->base_btf, s);
1723 if (off != -ENOENT)
1724 return off;
1725 }
1726
1727 /* BTF needs to be in a modifiable state to build string lookup index */
1728 if (btf_ensure_modifiable(btf))
1729 return libbpf_err(-ENOMEM);
1730
1731 off = strset__find_str(btf->strs_set, s);
1732 if (off < 0)
1733 return libbpf_err(off);
1734
1735 return btf->start_str_off + off;
1736 }
1737
1738 /* Add a string s to the BTF string section.
1739 * Returns:
1740 * - > 0 offset into string section, on success;
1741 * - < 0, on error.
1742 */
btf__add_str(struct btf * btf,const char * s)1743 int btf__add_str(struct btf *btf, const char *s)
1744 {
1745 int off;
1746
1747 if (btf->base_btf) {
1748 off = btf__find_str(btf->base_btf, s);
1749 if (off != -ENOENT)
1750 return off;
1751 }
1752
1753 if (btf_ensure_modifiable(btf))
1754 return libbpf_err(-ENOMEM);
1755
1756 off = strset__add_str(btf->strs_set, s);
1757 if (off < 0)
1758 return libbpf_err(off);
1759
1760 btf->hdr->str_len = strset__data_size(btf->strs_set);
1761
1762 return btf->start_str_off + off;
1763 }
1764
btf_add_type_mem(struct btf * btf,size_t add_sz)1765 static void *btf_add_type_mem(struct btf *btf, size_t add_sz)
1766 {
1767 return libbpf_add_mem(&btf->types_data, &btf->types_data_cap, 1,
1768 btf->hdr->type_len, UINT_MAX, add_sz);
1769 }
1770
btf_type_inc_vlen(struct btf_type * t)1771 static void btf_type_inc_vlen(struct btf_type *t)
1772 {
1773 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, btf_kflag(t));
1774 }
1775
btf_commit_type(struct btf * btf,int data_sz)1776 static int btf_commit_type(struct btf *btf, int data_sz)
1777 {
1778 int err;
1779
1780 err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1781 if (err)
1782 return libbpf_err(err);
1783
1784 btf->hdr->type_len += data_sz;
1785 btf->hdr->str_off += data_sz;
1786 btf->nr_types++;
1787 return btf->start_id + btf->nr_types - 1;
1788 }
1789
1790 struct btf_pipe {
1791 const struct btf *src;
1792 struct btf *dst;
1793 struct hashmap *str_off_map; /* map string offsets from src to dst */
1794 };
1795
btf_rewrite_str(struct btf_pipe * p,__u32 * str_off)1796 static int btf_rewrite_str(struct btf_pipe *p, __u32 *str_off)
1797 {
1798 long mapped_off;
1799 int off, err;
1800
1801 if (!*str_off) /* nothing to do for empty strings */
1802 return 0;
1803
1804 if (p->str_off_map &&
1805 hashmap__find(p->str_off_map, *str_off, &mapped_off)) {
1806 *str_off = mapped_off;
1807 return 0;
1808 }
1809
1810 off = btf__add_str(p->dst, btf__str_by_offset(p->src, *str_off));
1811 if (off < 0)
1812 return off;
1813
1814 /* Remember string mapping from src to dst. It avoids
1815 * performing expensive string comparisons.
1816 */
1817 if (p->str_off_map) {
1818 err = hashmap__append(p->str_off_map, *str_off, off);
1819 if (err)
1820 return err;
1821 }
1822
1823 *str_off = off;
1824 return 0;
1825 }
1826
btf_add_type(struct btf_pipe * p,const struct btf_type * src_type)1827 static int btf_add_type(struct btf_pipe *p, const struct btf_type *src_type)
1828 {
1829 struct btf_field_iter it;
1830 struct btf_type *t;
1831 __u32 *str_off;
1832 int sz, err;
1833
1834 sz = btf_type_size(src_type);
1835 if (sz < 0)
1836 return libbpf_err(sz);
1837
1838 /* deconstruct BTF, if necessary, and invalidate raw_data */
1839 if (btf_ensure_modifiable(p->dst))
1840 return libbpf_err(-ENOMEM);
1841
1842 t = btf_add_type_mem(p->dst, sz);
1843 if (!t)
1844 return libbpf_err(-ENOMEM);
1845
1846 memcpy(t, src_type, sz);
1847
1848 err = btf_field_iter_init(&it, t, BTF_FIELD_ITER_STRS);
1849 if (err)
1850 return libbpf_err(err);
1851
1852 while ((str_off = btf_field_iter_next(&it))) {
1853 err = btf_rewrite_str(p, str_off);
1854 if (err)
1855 return libbpf_err(err);
1856 }
1857
1858 return btf_commit_type(p->dst, sz);
1859 }
1860
btf__add_type(struct btf * btf,const struct btf * src_btf,const struct btf_type * src_type)1861 int btf__add_type(struct btf *btf, const struct btf *src_btf, const struct btf_type *src_type)
1862 {
1863 struct btf_pipe p = { .src = src_btf, .dst = btf };
1864
1865 return btf_add_type(&p, src_type);
1866 }
1867
1868 static size_t btf_dedup_identity_hash_fn(long key, void *ctx);
1869 static bool btf_dedup_equal_fn(long k1, long k2, void *ctx);
1870
btf__add_btf(struct btf * btf,const struct btf * src_btf)1871 int btf__add_btf(struct btf *btf, const struct btf *src_btf)
1872 {
1873 struct btf_pipe p = { .src = src_btf, .dst = btf };
1874 int data_sz, sz, cnt, i, err, old_strs_len;
1875 __u32 *off;
1876 void *t;
1877
1878 /* appending split BTF isn't supported yet */
1879 if (src_btf->base_btf)
1880 return libbpf_err(-ENOTSUP);
1881
1882 /* deconstruct BTF, if necessary, and invalidate raw_data */
1883 if (btf_ensure_modifiable(btf))
1884 return libbpf_err(-ENOMEM);
1885
1886 /* remember original strings section size if we have to roll back
1887 * partial strings section changes
1888 */
1889 old_strs_len = btf->hdr->str_len;
1890
1891 data_sz = src_btf->hdr->type_len;
1892 cnt = btf__type_cnt(src_btf) - 1;
1893
1894 /* pre-allocate enough memory for new types */
1895 t = btf_add_type_mem(btf, data_sz);
1896 if (!t)
1897 return libbpf_err(-ENOMEM);
1898
1899 /* pre-allocate enough memory for type offset index for new types */
1900 off = btf_add_type_offs_mem(btf, cnt);
1901 if (!off)
1902 return libbpf_err(-ENOMEM);
1903
1904 /* Map the string offsets from src_btf to the offsets from btf to improve performance */
1905 p.str_off_map = hashmap__new(btf_dedup_identity_hash_fn, btf_dedup_equal_fn, NULL);
1906 if (IS_ERR(p.str_off_map))
1907 return libbpf_err(-ENOMEM);
1908
1909 /* bulk copy types data for all types from src_btf */
1910 memcpy(t, src_btf->types_data, data_sz);
1911
1912 for (i = 0; i < cnt; i++) {
1913 struct btf_field_iter it;
1914 __u32 *type_id, *str_off;
1915
1916 sz = btf_type_size(t);
1917 if (sz < 0) {
1918 /* unlikely, has to be corrupted src_btf */
1919 err = sz;
1920 goto err_out;
1921 }
1922
1923 /* fill out type ID to type offset mapping for lookups by type ID */
1924 *off = t - btf->types_data;
1925
1926 /* add, dedup, and remap strings referenced by this BTF type */
1927 err = btf_field_iter_init(&it, t, BTF_FIELD_ITER_STRS);
1928 if (err)
1929 goto err_out;
1930 while ((str_off = btf_field_iter_next(&it))) {
1931 err = btf_rewrite_str(&p, str_off);
1932 if (err)
1933 goto err_out;
1934 }
1935
1936 /* remap all type IDs referenced from this BTF type */
1937 err = btf_field_iter_init(&it, t, BTF_FIELD_ITER_IDS);
1938 if (err)
1939 goto err_out;
1940
1941 while ((type_id = btf_field_iter_next(&it))) {
1942 if (!*type_id) /* nothing to do for VOID references */
1943 continue;
1944
1945 /* we haven't updated btf's type count yet, so
1946 * btf->start_id + btf->nr_types - 1 is the type ID offset we should
1947 * add to all newly added BTF types
1948 */
1949 *type_id += btf->start_id + btf->nr_types - 1;
1950 }
1951
1952 /* go to next type data and type offset index entry */
1953 t += sz;
1954 off++;
1955 }
1956
1957 /* Up until now any of the copied type data was effectively invisible,
1958 * so if we exited early before this point due to error, BTF would be
1959 * effectively unmodified. There would be extra internal memory
1960 * pre-allocated, but it would not be available for querying. But now
1961 * that we've copied and rewritten all the data successfully, we can
1962 * update type count and various internal offsets and sizes to
1963 * "commit" the changes and made them visible to the outside world.
1964 */
1965 btf->hdr->type_len += data_sz;
1966 btf->hdr->str_off += data_sz;
1967 btf->nr_types += cnt;
1968
1969 hashmap__free(p.str_off_map);
1970
1971 /* return type ID of the first added BTF type */
1972 return btf->start_id + btf->nr_types - cnt;
1973 err_out:
1974 /* zero out preallocated memory as if it was just allocated with
1975 * libbpf_add_mem()
1976 */
1977 memset(btf->types_data + btf->hdr->type_len, 0, data_sz);
1978 memset(btf->strs_data + old_strs_len, 0, btf->hdr->str_len - old_strs_len);
1979
1980 /* and now restore original strings section size; types data size
1981 * wasn't modified, so doesn't need restoring, see big comment above
1982 */
1983 btf->hdr->str_len = old_strs_len;
1984
1985 hashmap__free(p.str_off_map);
1986
1987 return libbpf_err(err);
1988 }
1989
1990 /*
1991 * Append new BTF_KIND_INT type with:
1992 * - *name* - non-empty, non-NULL type name;
1993 * - *sz* - power-of-2 (1, 2, 4, ..) size of the type, in bytes;
1994 * - encoding is a combination of BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL.
1995 * Returns:
1996 * - >0, type ID of newly added BTF type;
1997 * - <0, on error.
1998 */
btf__add_int(struct btf * btf,const char * name,size_t byte_sz,int encoding)1999 int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding)
2000 {
2001 struct btf_type *t;
2002 int sz, name_off;
2003
2004 /* non-empty name */
2005 if (!name || !name[0])
2006 return libbpf_err(-EINVAL);
2007 /* byte_sz must be power of 2 */
2008 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
2009 return libbpf_err(-EINVAL);
2010 if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
2011 return libbpf_err(-EINVAL);
2012
2013 /* deconstruct BTF, if necessary, and invalidate raw_data */
2014 if (btf_ensure_modifiable(btf))
2015 return libbpf_err(-ENOMEM);
2016
2017 sz = sizeof(struct btf_type) + sizeof(int);
2018 t = btf_add_type_mem(btf, sz);
2019 if (!t)
2020 return libbpf_err(-ENOMEM);
2021
2022 /* if something goes wrong later, we might end up with an extra string,
2023 * but that shouldn't be a problem, because BTF can't be constructed
2024 * completely anyway and will most probably be just discarded
2025 */
2026 name_off = btf__add_str(btf, name);
2027 if (name_off < 0)
2028 return name_off;
2029
2030 t->name_off = name_off;
2031 t->info = btf_type_info(BTF_KIND_INT, 0, 0);
2032 t->size = byte_sz;
2033 /* set INT info, we don't allow setting legacy bit offset/size */
2034 *(__u32 *)(t + 1) = (encoding << 24) | (byte_sz * 8);
2035
2036 return btf_commit_type(btf, sz);
2037 }
2038
2039 /*
2040 * Append new BTF_KIND_FLOAT type with:
2041 * - *name* - non-empty, non-NULL type name;
2042 * - *sz* - size of the type, in bytes;
2043 * Returns:
2044 * - >0, type ID of newly added BTF type;
2045 * - <0, on error.
2046 */
btf__add_float(struct btf * btf,const char * name,size_t byte_sz)2047 int btf__add_float(struct btf *btf, const char *name, size_t byte_sz)
2048 {
2049 struct btf_type *t;
2050 int sz, name_off;
2051
2052 /* non-empty name */
2053 if (!name || !name[0])
2054 return libbpf_err(-EINVAL);
2055
2056 /* byte_sz must be one of the explicitly allowed values */
2057 if (byte_sz != 2 && byte_sz != 4 && byte_sz != 8 && byte_sz != 12 &&
2058 byte_sz != 16)
2059 return libbpf_err(-EINVAL);
2060
2061 if (btf_ensure_modifiable(btf))
2062 return libbpf_err(-ENOMEM);
2063
2064 sz = sizeof(struct btf_type);
2065 t = btf_add_type_mem(btf, sz);
2066 if (!t)
2067 return libbpf_err(-ENOMEM);
2068
2069 name_off = btf__add_str(btf, name);
2070 if (name_off < 0)
2071 return name_off;
2072
2073 t->name_off = name_off;
2074 t->info = btf_type_info(BTF_KIND_FLOAT, 0, 0);
2075 t->size = byte_sz;
2076
2077 return btf_commit_type(btf, sz);
2078 }
2079
2080 /* it's completely legal to append BTF types with type IDs pointing forward to
2081 * types that haven't been appended yet, so we only make sure that id looks
2082 * sane, we can't guarantee that ID will always be valid
2083 */
validate_type_id(int id)2084 static int validate_type_id(int id)
2085 {
2086 if (id < 0 || id > BTF_MAX_NR_TYPES)
2087 return -EINVAL;
2088 return 0;
2089 }
2090
2091 /* generic append function for PTR, TYPEDEF, CONST/VOLATILE/RESTRICT */
btf_add_ref_kind(struct btf * btf,int kind,const char * name,int ref_type_id)2092 static int btf_add_ref_kind(struct btf *btf, int kind, const char *name, int ref_type_id)
2093 {
2094 struct btf_type *t;
2095 int sz, name_off = 0;
2096
2097 if (validate_type_id(ref_type_id))
2098 return libbpf_err(-EINVAL);
2099
2100 if (btf_ensure_modifiable(btf))
2101 return libbpf_err(-ENOMEM);
2102
2103 sz = sizeof(struct btf_type);
2104 t = btf_add_type_mem(btf, sz);
2105 if (!t)
2106 return libbpf_err(-ENOMEM);
2107
2108 if (name && name[0]) {
2109 name_off = btf__add_str(btf, name);
2110 if (name_off < 0)
2111 return name_off;
2112 }
2113
2114 t->name_off = name_off;
2115 t->info = btf_type_info(kind, 0, 0);
2116 t->type = ref_type_id;
2117
2118 return btf_commit_type(btf, sz);
2119 }
2120
2121 /*
2122 * Append new BTF_KIND_PTR type with:
2123 * - *ref_type_id* - referenced type ID, it might not exist yet;
2124 * Returns:
2125 * - >0, type ID of newly added BTF type;
2126 * - <0, on error.
2127 */
btf__add_ptr(struct btf * btf,int ref_type_id)2128 int btf__add_ptr(struct btf *btf, int ref_type_id)
2129 {
2130 return btf_add_ref_kind(btf, BTF_KIND_PTR, NULL, ref_type_id);
2131 }
2132
2133 /*
2134 * Append new BTF_KIND_ARRAY type with:
2135 * - *index_type_id* - type ID of the type describing array index;
2136 * - *elem_type_id* - type ID of the type describing array element;
2137 * - *nr_elems* - the size of the array;
2138 * Returns:
2139 * - >0, type ID of newly added BTF type;
2140 * - <0, on error.
2141 */
btf__add_array(struct btf * btf,int index_type_id,int elem_type_id,__u32 nr_elems)2142 int btf__add_array(struct btf *btf, int index_type_id, int elem_type_id, __u32 nr_elems)
2143 {
2144 struct btf_type *t;
2145 struct btf_array *a;
2146 int sz;
2147
2148 if (validate_type_id(index_type_id) || validate_type_id(elem_type_id))
2149 return libbpf_err(-EINVAL);
2150
2151 if (btf_ensure_modifiable(btf))
2152 return libbpf_err(-ENOMEM);
2153
2154 sz = sizeof(struct btf_type) + sizeof(struct btf_array);
2155 t = btf_add_type_mem(btf, sz);
2156 if (!t)
2157 return libbpf_err(-ENOMEM);
2158
2159 t->name_off = 0;
2160 t->info = btf_type_info(BTF_KIND_ARRAY, 0, 0);
2161 t->size = 0;
2162
2163 a = btf_array(t);
2164 a->type = elem_type_id;
2165 a->index_type = index_type_id;
2166 a->nelems = nr_elems;
2167
2168 return btf_commit_type(btf, sz);
2169 }
2170
2171 /* generic STRUCT/UNION append function */
btf_add_composite(struct btf * btf,int kind,const char * name,__u32 bytes_sz)2172 static int btf_add_composite(struct btf *btf, int kind, const char *name, __u32 bytes_sz)
2173 {
2174 struct btf_type *t;
2175 int sz, name_off = 0;
2176
2177 if (btf_ensure_modifiable(btf))
2178 return libbpf_err(-ENOMEM);
2179
2180 sz = sizeof(struct btf_type);
2181 t = btf_add_type_mem(btf, sz);
2182 if (!t)
2183 return libbpf_err(-ENOMEM);
2184
2185 if (name && name[0]) {
2186 name_off = btf__add_str(btf, name);
2187 if (name_off < 0)
2188 return name_off;
2189 }
2190
2191 /* start out with vlen=0 and no kflag; this will be adjusted when
2192 * adding each member
2193 */
2194 t->name_off = name_off;
2195 t->info = btf_type_info(kind, 0, 0);
2196 t->size = bytes_sz;
2197
2198 return btf_commit_type(btf, sz);
2199 }
2200
2201 /*
2202 * Append new BTF_KIND_STRUCT type with:
2203 * - *name* - name of the struct, can be NULL or empty for anonymous structs;
2204 * - *byte_sz* - size of the struct, in bytes;
2205 *
2206 * Struct initially has no fields in it. Fields can be added by
2207 * btf__add_field() right after btf__add_struct() succeeds.
2208 *
2209 * Returns:
2210 * - >0, type ID of newly added BTF type;
2211 * - <0, on error.
2212 */
btf__add_struct(struct btf * btf,const char * name,__u32 byte_sz)2213 int btf__add_struct(struct btf *btf, const char *name, __u32 byte_sz)
2214 {
2215 return btf_add_composite(btf, BTF_KIND_STRUCT, name, byte_sz);
2216 }
2217
2218 /*
2219 * Append new BTF_KIND_UNION type with:
2220 * - *name* - name of the union, can be NULL or empty for anonymous union;
2221 * - *byte_sz* - size of the union, in bytes;
2222 *
2223 * Union initially has no fields in it. Fields can be added by
2224 * btf__add_field() right after btf__add_union() succeeds. All fields
2225 * should have *bit_offset* of 0.
2226 *
2227 * Returns:
2228 * - >0, type ID of newly added BTF type;
2229 * - <0, on error.
2230 */
btf__add_union(struct btf * btf,const char * name,__u32 byte_sz)2231 int btf__add_union(struct btf *btf, const char *name, __u32 byte_sz)
2232 {
2233 return btf_add_composite(btf, BTF_KIND_UNION, name, byte_sz);
2234 }
2235
btf_last_type(struct btf * btf)2236 static struct btf_type *btf_last_type(struct btf *btf)
2237 {
2238 return btf_type_by_id(btf, btf__type_cnt(btf) - 1);
2239 }
2240
2241 /*
2242 * Append new field for the current STRUCT/UNION type with:
2243 * - *name* - name of the field, can be NULL or empty for anonymous field;
2244 * - *type_id* - type ID for the type describing field type;
2245 * - *bit_offset* - bit offset of the start of the field within struct/union;
2246 * - *bit_size* - bit size of a bitfield, 0 for non-bitfield fields;
2247 * Returns:
2248 * - 0, on success;
2249 * - <0, on error.
2250 */
btf__add_field(struct btf * btf,const char * name,int type_id,__u32 bit_offset,__u32 bit_size)2251 int btf__add_field(struct btf *btf, const char *name, int type_id,
2252 __u32 bit_offset, __u32 bit_size)
2253 {
2254 struct btf_type *t;
2255 struct btf_member *m;
2256 bool is_bitfield;
2257 int sz, name_off = 0;
2258
2259 /* last type should be union/struct */
2260 if (btf->nr_types == 0)
2261 return libbpf_err(-EINVAL);
2262 t = btf_last_type(btf);
2263 if (!btf_is_composite(t))
2264 return libbpf_err(-EINVAL);
2265
2266 if (validate_type_id(type_id))
2267 return libbpf_err(-EINVAL);
2268 /* best-effort bit field offset/size enforcement */
2269 is_bitfield = bit_size || (bit_offset % 8 != 0);
2270 if (is_bitfield && (bit_size == 0 || bit_size > 255 || bit_offset > 0xffffff))
2271 return libbpf_err(-EINVAL);
2272
2273 /* only offset 0 is allowed for unions */
2274 if (btf_is_union(t) && bit_offset)
2275 return libbpf_err(-EINVAL);
2276
2277 /* decompose and invalidate raw data */
2278 if (btf_ensure_modifiable(btf))
2279 return libbpf_err(-ENOMEM);
2280
2281 sz = sizeof(struct btf_member);
2282 m = btf_add_type_mem(btf, sz);
2283 if (!m)
2284 return libbpf_err(-ENOMEM);
2285
2286 if (name && name[0]) {
2287 name_off = btf__add_str(btf, name);
2288 if (name_off < 0)
2289 return name_off;
2290 }
2291
2292 m->name_off = name_off;
2293 m->type = type_id;
2294 m->offset = bit_offset | (bit_size << 24);
2295
2296 /* btf_add_type_mem can invalidate t pointer */
2297 t = btf_last_type(btf);
2298 /* update parent type's vlen and kflag */
2299 t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, is_bitfield || btf_kflag(t));
2300
2301 btf->hdr->type_len += sz;
2302 btf->hdr->str_off += sz;
2303 return 0;
2304 }
2305
btf_add_enum_common(struct btf * btf,const char * name,__u32 byte_sz,bool is_signed,__u8 kind)2306 static int btf_add_enum_common(struct btf *btf, const char *name, __u32 byte_sz,
2307 bool is_signed, __u8 kind)
2308 {
2309 struct btf_type *t;
2310 int sz, name_off = 0;
2311
2312 /* byte_sz must be power of 2 */
2313 if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 8)
2314 return libbpf_err(-EINVAL);
2315
2316 if (btf_ensure_modifiable(btf))
2317 return libbpf_err(-ENOMEM);
2318
2319 sz = sizeof(struct btf_type);
2320 t = btf_add_type_mem(btf, sz);
2321 if (!t)
2322 return libbpf_err(-ENOMEM);
2323
2324 if (name && name[0]) {
2325 name_off = btf__add_str(btf, name);
2326 if (name_off < 0)
2327 return name_off;
2328 }
2329
2330 /* start out with vlen=0; it will be adjusted when adding enum values */
2331 t->name_off = name_off;
2332 t->info = btf_type_info(kind, 0, is_signed);
2333 t->size = byte_sz;
2334
2335 return btf_commit_type(btf, sz);
2336 }
2337
2338 /*
2339 * Append new BTF_KIND_ENUM type with:
2340 * - *name* - name of the enum, can be NULL or empty for anonymous enums;
2341 * - *byte_sz* - size of the enum, in bytes.
2342 *
2343 * Enum initially has no enum values in it (and corresponds to enum forward
2344 * declaration). Enumerator values can be added by btf__add_enum_value()
2345 * immediately after btf__add_enum() succeeds.
2346 *
2347 * Returns:
2348 * - >0, type ID of newly added BTF type;
2349 * - <0, on error.
2350 */
btf__add_enum(struct btf * btf,const char * name,__u32 byte_sz)2351 int btf__add_enum(struct btf *btf, const char *name, __u32 byte_sz)
2352 {
2353 /*
2354 * set the signedness to be unsigned, it will change to signed
2355 * if any later enumerator is negative.
2356 */
2357 return btf_add_enum_common(btf, name, byte_sz, false, BTF_KIND_ENUM);
2358 }
2359
2360 /*
2361 * Append new enum value for the current ENUM type with:
2362 * - *name* - name of the enumerator value, can't be NULL or empty;
2363 * - *value* - integer value corresponding to enum value *name*;
2364 * Returns:
2365 * - 0, on success;
2366 * - <0, on error.
2367 */
btf__add_enum_value(struct btf * btf,const char * name,__s64 value)2368 int btf__add_enum_value(struct btf *btf, const char *name, __s64 value)
2369 {
2370 struct btf_type *t;
2371 struct btf_enum *v;
2372 int sz, name_off;
2373
2374 /* last type should be BTF_KIND_ENUM */
2375 if (btf->nr_types == 0)
2376 return libbpf_err(-EINVAL);
2377 t = btf_last_type(btf);
2378 if (!btf_is_enum(t))
2379 return libbpf_err(-EINVAL);
2380
2381 /* non-empty name */
2382 if (!name || !name[0])
2383 return libbpf_err(-EINVAL);
2384 if (value < INT_MIN || value > UINT_MAX)
2385 return libbpf_err(-E2BIG);
2386
2387 /* decompose and invalidate raw data */
2388 if (btf_ensure_modifiable(btf))
2389 return libbpf_err(-ENOMEM);
2390
2391 sz = sizeof(struct btf_enum);
2392 v = btf_add_type_mem(btf, sz);
2393 if (!v)
2394 return libbpf_err(-ENOMEM);
2395
2396 name_off = btf__add_str(btf, name);
2397 if (name_off < 0)
2398 return name_off;
2399
2400 v->name_off = name_off;
2401 v->val = value;
2402
2403 /* update parent type's vlen */
2404 t = btf_last_type(btf);
2405 btf_type_inc_vlen(t);
2406
2407 /* if negative value, set signedness to signed */
2408 if (value < 0)
2409 t->info = btf_type_info(btf_kind(t), btf_vlen(t), true);
2410
2411 btf->hdr->type_len += sz;
2412 btf->hdr->str_off += sz;
2413 return 0;
2414 }
2415
2416 /*
2417 * Append new BTF_KIND_ENUM64 type with:
2418 * - *name* - name of the enum, can be NULL or empty for anonymous enums;
2419 * - *byte_sz* - size of the enum, in bytes.
2420 * - *is_signed* - whether the enum values are signed or not;
2421 *
2422 * Enum initially has no enum values in it (and corresponds to enum forward
2423 * declaration). Enumerator values can be added by btf__add_enum64_value()
2424 * immediately after btf__add_enum64() succeeds.
2425 *
2426 * Returns:
2427 * - >0, type ID of newly added BTF type;
2428 * - <0, on error.
2429 */
btf__add_enum64(struct btf * btf,const char * name,__u32 byte_sz,bool is_signed)2430 int btf__add_enum64(struct btf *btf, const char *name, __u32 byte_sz,
2431 bool is_signed)
2432 {
2433 return btf_add_enum_common(btf, name, byte_sz, is_signed,
2434 BTF_KIND_ENUM64);
2435 }
2436
2437 /*
2438 * Append new enum value for the current ENUM64 type with:
2439 * - *name* - name of the enumerator value, can't be NULL or empty;
2440 * - *value* - integer value corresponding to enum value *name*;
2441 * Returns:
2442 * - 0, on success;
2443 * - <0, on error.
2444 */
btf__add_enum64_value(struct btf * btf,const char * name,__u64 value)2445 int btf__add_enum64_value(struct btf *btf, const char *name, __u64 value)
2446 {
2447 struct btf_enum64 *v;
2448 struct btf_type *t;
2449 int sz, name_off;
2450
2451 /* last type should be BTF_KIND_ENUM64 */
2452 if (btf->nr_types == 0)
2453 return libbpf_err(-EINVAL);
2454 t = btf_last_type(btf);
2455 if (!btf_is_enum64(t))
2456 return libbpf_err(-EINVAL);
2457
2458 /* non-empty name */
2459 if (!name || !name[0])
2460 return libbpf_err(-EINVAL);
2461
2462 /* decompose and invalidate raw data */
2463 if (btf_ensure_modifiable(btf))
2464 return libbpf_err(-ENOMEM);
2465
2466 sz = sizeof(struct btf_enum64);
2467 v = btf_add_type_mem(btf, sz);
2468 if (!v)
2469 return libbpf_err(-ENOMEM);
2470
2471 name_off = btf__add_str(btf, name);
2472 if (name_off < 0)
2473 return name_off;
2474
2475 v->name_off = name_off;
2476 v->val_lo32 = (__u32)value;
2477 v->val_hi32 = value >> 32;
2478
2479 /* update parent type's vlen */
2480 t = btf_last_type(btf);
2481 btf_type_inc_vlen(t);
2482
2483 btf->hdr->type_len += sz;
2484 btf->hdr->str_off += sz;
2485 return 0;
2486 }
2487
2488 /*
2489 * Append new BTF_KIND_FWD type with:
2490 * - *name*, non-empty/non-NULL name;
2491 * - *fwd_kind*, kind of forward declaration, one of BTF_FWD_STRUCT,
2492 * BTF_FWD_UNION, or BTF_FWD_ENUM;
2493 * Returns:
2494 * - >0, type ID of newly added BTF type;
2495 * - <0, on error.
2496 */
btf__add_fwd(struct btf * btf,const char * name,enum btf_fwd_kind fwd_kind)2497 int btf__add_fwd(struct btf *btf, const char *name, enum btf_fwd_kind fwd_kind)
2498 {
2499 if (!name || !name[0])
2500 return libbpf_err(-EINVAL);
2501
2502 switch (fwd_kind) {
2503 case BTF_FWD_STRUCT:
2504 case BTF_FWD_UNION: {
2505 struct btf_type *t;
2506 int id;
2507
2508 id = btf_add_ref_kind(btf, BTF_KIND_FWD, name, 0);
2509 if (id <= 0)
2510 return id;
2511 t = btf_type_by_id(btf, id);
2512 t->info = btf_type_info(BTF_KIND_FWD, 0, fwd_kind == BTF_FWD_UNION);
2513 return id;
2514 }
2515 case BTF_FWD_ENUM:
2516 /* enum forward in BTF currently is just an enum with no enum
2517 * values; we also assume a standard 4-byte size for it
2518 */
2519 return btf__add_enum(btf, name, sizeof(int));
2520 default:
2521 return libbpf_err(-EINVAL);
2522 }
2523 }
2524
2525 /*
2526 * Append new BTF_KING_TYPEDEF type with:
2527 * - *name*, non-empty/non-NULL name;
2528 * - *ref_type_id* - referenced type ID, it might not exist yet;
2529 * Returns:
2530 * - >0, type ID of newly added BTF type;
2531 * - <0, on error.
2532 */
btf__add_typedef(struct btf * btf,const char * name,int ref_type_id)2533 int btf__add_typedef(struct btf *btf, const char *name, int ref_type_id)
2534 {
2535 if (!name || !name[0])
2536 return libbpf_err(-EINVAL);
2537
2538 return btf_add_ref_kind(btf, BTF_KIND_TYPEDEF, name, ref_type_id);
2539 }
2540
2541 /*
2542 * Append new BTF_KIND_VOLATILE type with:
2543 * - *ref_type_id* - referenced type ID, it might not exist yet;
2544 * Returns:
2545 * - >0, type ID of newly added BTF type;
2546 * - <0, on error.
2547 */
btf__add_volatile(struct btf * btf,int ref_type_id)2548 int btf__add_volatile(struct btf *btf, int ref_type_id)
2549 {
2550 return btf_add_ref_kind(btf, BTF_KIND_VOLATILE, NULL, ref_type_id);
2551 }
2552
2553 /*
2554 * Append new BTF_KIND_CONST type with:
2555 * - *ref_type_id* - referenced type ID, it might not exist yet;
2556 * Returns:
2557 * - >0, type ID of newly added BTF type;
2558 * - <0, on error.
2559 */
btf__add_const(struct btf * btf,int ref_type_id)2560 int btf__add_const(struct btf *btf, int ref_type_id)
2561 {
2562 return btf_add_ref_kind(btf, BTF_KIND_CONST, NULL, ref_type_id);
2563 }
2564
2565 /*
2566 * Append new BTF_KIND_RESTRICT type with:
2567 * - *ref_type_id* - referenced type ID, it might not exist yet;
2568 * Returns:
2569 * - >0, type ID of newly added BTF type;
2570 * - <0, on error.
2571 */
btf__add_restrict(struct btf * btf,int ref_type_id)2572 int btf__add_restrict(struct btf *btf, int ref_type_id)
2573 {
2574 return btf_add_ref_kind(btf, BTF_KIND_RESTRICT, NULL, ref_type_id);
2575 }
2576
2577 /*
2578 * Append new BTF_KIND_TYPE_TAG type with:
2579 * - *value*, non-empty/non-NULL tag value;
2580 * - *ref_type_id* - referenced type ID, it might not exist yet;
2581 * Returns:
2582 * - >0, type ID of newly added BTF type;
2583 * - <0, on error.
2584 */
btf__add_type_tag(struct btf * btf,const char * value,int ref_type_id)2585 int btf__add_type_tag(struct btf *btf, const char *value, int ref_type_id)
2586 {
2587 if (!value || !value[0])
2588 return libbpf_err(-EINVAL);
2589
2590 return btf_add_ref_kind(btf, BTF_KIND_TYPE_TAG, value, ref_type_id);
2591 }
2592
2593 /*
2594 * Append new BTF_KIND_FUNC type with:
2595 * - *name*, non-empty/non-NULL name;
2596 * - *proto_type_id* - FUNC_PROTO's type ID, it might not exist yet;
2597 * Returns:
2598 * - >0, type ID of newly added BTF type;
2599 * - <0, on error.
2600 */
btf__add_func(struct btf * btf,const char * name,enum btf_func_linkage linkage,int proto_type_id)2601 int btf__add_func(struct btf *btf, const char *name,
2602 enum btf_func_linkage linkage, int proto_type_id)
2603 {
2604 int id;
2605
2606 if (!name || !name[0])
2607 return libbpf_err(-EINVAL);
2608 if (linkage != BTF_FUNC_STATIC && linkage != BTF_FUNC_GLOBAL &&
2609 linkage != BTF_FUNC_EXTERN)
2610 return libbpf_err(-EINVAL);
2611
2612 id = btf_add_ref_kind(btf, BTF_KIND_FUNC, name, proto_type_id);
2613 if (id > 0) {
2614 struct btf_type *t = btf_type_by_id(btf, id);
2615
2616 t->info = btf_type_info(BTF_KIND_FUNC, linkage, 0);
2617 }
2618 return libbpf_err(id);
2619 }
2620
2621 /*
2622 * Append new BTF_KIND_FUNC_PROTO with:
2623 * - *ret_type_id* - type ID for return result of a function.
2624 *
2625 * Function prototype initially has no arguments, but they can be added by
2626 * btf__add_func_param() one by one, immediately after
2627 * btf__add_func_proto() succeeded.
2628 *
2629 * Returns:
2630 * - >0, type ID of newly added BTF type;
2631 * - <0, on error.
2632 */
btf__add_func_proto(struct btf * btf,int ret_type_id)2633 int btf__add_func_proto(struct btf *btf, int ret_type_id)
2634 {
2635 struct btf_type *t;
2636 int sz;
2637
2638 if (validate_type_id(ret_type_id))
2639 return libbpf_err(-EINVAL);
2640
2641 if (btf_ensure_modifiable(btf))
2642 return libbpf_err(-ENOMEM);
2643
2644 sz = sizeof(struct btf_type);
2645 t = btf_add_type_mem(btf, sz);
2646 if (!t)
2647 return libbpf_err(-ENOMEM);
2648
2649 /* start out with vlen=0; this will be adjusted when adding enum
2650 * values, if necessary
2651 */
2652 t->name_off = 0;
2653 t->info = btf_type_info(BTF_KIND_FUNC_PROTO, 0, 0);
2654 t->type = ret_type_id;
2655
2656 return btf_commit_type(btf, sz);
2657 }
2658
2659 /*
2660 * Append new function parameter for current FUNC_PROTO type with:
2661 * - *name* - parameter name, can be NULL or empty;
2662 * - *type_id* - type ID describing the type of the parameter.
2663 * Returns:
2664 * - 0, on success;
2665 * - <0, on error.
2666 */
btf__add_func_param(struct btf * btf,const char * name,int type_id)2667 int btf__add_func_param(struct btf *btf, const char *name, int type_id)
2668 {
2669 struct btf_type *t;
2670 struct btf_param *p;
2671 int sz, name_off = 0;
2672
2673 if (validate_type_id(type_id))
2674 return libbpf_err(-EINVAL);
2675
2676 /* last type should be BTF_KIND_FUNC_PROTO */
2677 if (btf->nr_types == 0)
2678 return libbpf_err(-EINVAL);
2679 t = btf_last_type(btf);
2680 if (!btf_is_func_proto(t))
2681 return libbpf_err(-EINVAL);
2682
2683 /* decompose and invalidate raw data */
2684 if (btf_ensure_modifiable(btf))
2685 return libbpf_err(-ENOMEM);
2686
2687 sz = sizeof(struct btf_param);
2688 p = btf_add_type_mem(btf, sz);
2689 if (!p)
2690 return libbpf_err(-ENOMEM);
2691
2692 if (name && name[0]) {
2693 name_off = btf__add_str(btf, name);
2694 if (name_off < 0)
2695 return name_off;
2696 }
2697
2698 p->name_off = name_off;
2699 p->type = type_id;
2700
2701 /* update parent type's vlen */
2702 t = btf_last_type(btf);
2703 btf_type_inc_vlen(t);
2704
2705 btf->hdr->type_len += sz;
2706 btf->hdr->str_off += sz;
2707 return 0;
2708 }
2709
2710 /*
2711 * Append new BTF_KIND_VAR type with:
2712 * - *name* - non-empty/non-NULL name;
2713 * - *linkage* - variable linkage, one of BTF_VAR_STATIC,
2714 * BTF_VAR_GLOBAL_ALLOCATED, or BTF_VAR_GLOBAL_EXTERN;
2715 * - *type_id* - type ID of the type describing the type of the variable.
2716 * Returns:
2717 * - >0, type ID of newly added BTF type;
2718 * - <0, on error.
2719 */
btf__add_var(struct btf * btf,const char * name,int linkage,int type_id)2720 int btf__add_var(struct btf *btf, const char *name, int linkage, int type_id)
2721 {
2722 struct btf_type *t;
2723 struct btf_var *v;
2724 int sz, name_off;
2725
2726 /* non-empty name */
2727 if (!name || !name[0])
2728 return libbpf_err(-EINVAL);
2729 if (linkage != BTF_VAR_STATIC && linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2730 linkage != BTF_VAR_GLOBAL_EXTERN)
2731 return libbpf_err(-EINVAL);
2732 if (validate_type_id(type_id))
2733 return libbpf_err(-EINVAL);
2734
2735 /* deconstruct BTF, if necessary, and invalidate raw_data */
2736 if (btf_ensure_modifiable(btf))
2737 return libbpf_err(-ENOMEM);
2738
2739 sz = sizeof(struct btf_type) + sizeof(struct btf_var);
2740 t = btf_add_type_mem(btf, sz);
2741 if (!t)
2742 return libbpf_err(-ENOMEM);
2743
2744 name_off = btf__add_str(btf, name);
2745 if (name_off < 0)
2746 return name_off;
2747
2748 t->name_off = name_off;
2749 t->info = btf_type_info(BTF_KIND_VAR, 0, 0);
2750 t->type = type_id;
2751
2752 v = btf_var(t);
2753 v->linkage = linkage;
2754
2755 return btf_commit_type(btf, sz);
2756 }
2757
2758 /*
2759 * Append new BTF_KIND_DATASEC type with:
2760 * - *name* - non-empty/non-NULL name;
2761 * - *byte_sz* - data section size, in bytes.
2762 *
2763 * Data section is initially empty. Variables info can be added with
2764 * btf__add_datasec_var_info() calls, after btf__add_datasec() succeeds.
2765 *
2766 * Returns:
2767 * - >0, type ID of newly added BTF type;
2768 * - <0, on error.
2769 */
btf__add_datasec(struct btf * btf,const char * name,__u32 byte_sz)2770 int btf__add_datasec(struct btf *btf, const char *name, __u32 byte_sz)
2771 {
2772 struct btf_type *t;
2773 int sz, name_off;
2774
2775 /* non-empty name */
2776 if (!name || !name[0])
2777 return libbpf_err(-EINVAL);
2778
2779 if (btf_ensure_modifiable(btf))
2780 return libbpf_err(-ENOMEM);
2781
2782 sz = sizeof(struct btf_type);
2783 t = btf_add_type_mem(btf, sz);
2784 if (!t)
2785 return libbpf_err(-ENOMEM);
2786
2787 name_off = btf__add_str(btf, name);
2788 if (name_off < 0)
2789 return name_off;
2790
2791 /* start with vlen=0, which will be update as var_secinfos are added */
2792 t->name_off = name_off;
2793 t->info = btf_type_info(BTF_KIND_DATASEC, 0, 0);
2794 t->size = byte_sz;
2795
2796 return btf_commit_type(btf, sz);
2797 }
2798
2799 /*
2800 * Append new data section variable information entry for current DATASEC type:
2801 * - *var_type_id* - type ID, describing type of the variable;
2802 * - *offset* - variable offset within data section, in bytes;
2803 * - *byte_sz* - variable size, in bytes.
2804 *
2805 * Returns:
2806 * - 0, on success;
2807 * - <0, on error.
2808 */
btf__add_datasec_var_info(struct btf * btf,int var_type_id,__u32 offset,__u32 byte_sz)2809 int btf__add_datasec_var_info(struct btf *btf, int var_type_id, __u32 offset, __u32 byte_sz)
2810 {
2811 struct btf_type *t;
2812 struct btf_var_secinfo *v;
2813 int sz;
2814
2815 /* last type should be BTF_KIND_DATASEC */
2816 if (btf->nr_types == 0)
2817 return libbpf_err(-EINVAL);
2818 t = btf_last_type(btf);
2819 if (!btf_is_datasec(t))
2820 return libbpf_err(-EINVAL);
2821
2822 if (validate_type_id(var_type_id))
2823 return libbpf_err(-EINVAL);
2824
2825 /* decompose and invalidate raw data */
2826 if (btf_ensure_modifiable(btf))
2827 return libbpf_err(-ENOMEM);
2828
2829 sz = sizeof(struct btf_var_secinfo);
2830 v = btf_add_type_mem(btf, sz);
2831 if (!v)
2832 return libbpf_err(-ENOMEM);
2833
2834 v->type = var_type_id;
2835 v->offset = offset;
2836 v->size = byte_sz;
2837
2838 /* update parent type's vlen */
2839 t = btf_last_type(btf);
2840 btf_type_inc_vlen(t);
2841
2842 btf->hdr->type_len += sz;
2843 btf->hdr->str_off += sz;
2844 return 0;
2845 }
2846
2847 /*
2848 * Append new BTF_KIND_DECL_TAG type with:
2849 * - *value* - non-empty/non-NULL string;
2850 * - *ref_type_id* - referenced type ID, it might not exist yet;
2851 * - *component_idx* - -1 for tagging reference type, otherwise struct/union
2852 * member or function argument index;
2853 * Returns:
2854 * - >0, type ID of newly added BTF type;
2855 * - <0, on error.
2856 */
btf__add_decl_tag(struct btf * btf,const char * value,int ref_type_id,int component_idx)2857 int btf__add_decl_tag(struct btf *btf, const char *value, int ref_type_id,
2858 int component_idx)
2859 {
2860 struct btf_type *t;
2861 int sz, value_off;
2862
2863 if (!value || !value[0] || component_idx < -1)
2864 return libbpf_err(-EINVAL);
2865
2866 if (validate_type_id(ref_type_id))
2867 return libbpf_err(-EINVAL);
2868
2869 if (btf_ensure_modifiable(btf))
2870 return libbpf_err(-ENOMEM);
2871
2872 sz = sizeof(struct btf_type) + sizeof(struct btf_decl_tag);
2873 t = btf_add_type_mem(btf, sz);
2874 if (!t)
2875 return libbpf_err(-ENOMEM);
2876
2877 value_off = btf__add_str(btf, value);
2878 if (value_off < 0)
2879 return value_off;
2880
2881 t->name_off = value_off;
2882 t->info = btf_type_info(BTF_KIND_DECL_TAG, 0, false);
2883 t->type = ref_type_id;
2884 btf_decl_tag(t)->component_idx = component_idx;
2885
2886 return btf_commit_type(btf, sz);
2887 }
2888
2889 struct btf_ext_sec_setup_param {
2890 __u32 off;
2891 __u32 len;
2892 __u32 min_rec_size;
2893 struct btf_ext_info *ext_info;
2894 const char *desc;
2895 };
2896
btf_ext_setup_info(struct btf_ext * btf_ext,struct btf_ext_sec_setup_param * ext_sec)2897 static int btf_ext_setup_info(struct btf_ext *btf_ext,
2898 struct btf_ext_sec_setup_param *ext_sec)
2899 {
2900 const struct btf_ext_info_sec *sinfo;
2901 struct btf_ext_info *ext_info;
2902 __u32 info_left, record_size;
2903 size_t sec_cnt = 0;
2904 /* The start of the info sec (including the __u32 record_size). */
2905 void *info;
2906
2907 if (ext_sec->len == 0)
2908 return 0;
2909
2910 if (ext_sec->off & 0x03) {
2911 pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
2912 ext_sec->desc);
2913 return -EINVAL;
2914 }
2915
2916 info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
2917 info_left = ext_sec->len;
2918
2919 if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
2920 pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
2921 ext_sec->desc, ext_sec->off, ext_sec->len);
2922 return -EINVAL;
2923 }
2924
2925 /* At least a record size */
2926 if (info_left < sizeof(__u32)) {
2927 pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
2928 return -EINVAL;
2929 }
2930
2931 /* The record size needs to meet the minimum standard */
2932 record_size = *(__u32 *)info;
2933 if (record_size < ext_sec->min_rec_size ||
2934 record_size & 0x03) {
2935 pr_debug("%s section in .BTF.ext has invalid record size %u\n",
2936 ext_sec->desc, record_size);
2937 return -EINVAL;
2938 }
2939
2940 sinfo = info + sizeof(__u32);
2941 info_left -= sizeof(__u32);
2942
2943 /* If no records, return failure now so .BTF.ext won't be used. */
2944 if (!info_left) {
2945 pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
2946 return -EINVAL;
2947 }
2948
2949 while (info_left) {
2950 unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
2951 __u64 total_record_size;
2952 __u32 num_records;
2953
2954 if (info_left < sec_hdrlen) {
2955 pr_debug("%s section header is not found in .BTF.ext\n",
2956 ext_sec->desc);
2957 return -EINVAL;
2958 }
2959
2960 num_records = sinfo->num_info;
2961 if (num_records == 0) {
2962 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2963 ext_sec->desc);
2964 return -EINVAL;
2965 }
2966
2967 total_record_size = sec_hdrlen + (__u64)num_records * record_size;
2968 if (info_left < total_record_size) {
2969 pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2970 ext_sec->desc);
2971 return -EINVAL;
2972 }
2973
2974 info_left -= total_record_size;
2975 sinfo = (void *)sinfo + total_record_size;
2976 sec_cnt++;
2977 }
2978
2979 ext_info = ext_sec->ext_info;
2980 ext_info->len = ext_sec->len - sizeof(__u32);
2981 ext_info->rec_size = record_size;
2982 ext_info->info = info + sizeof(__u32);
2983 ext_info->sec_cnt = sec_cnt;
2984
2985 return 0;
2986 }
2987
btf_ext_setup_func_info(struct btf_ext * btf_ext)2988 static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
2989 {
2990 struct btf_ext_sec_setup_param param = {
2991 .off = btf_ext->hdr->func_info_off,
2992 .len = btf_ext->hdr->func_info_len,
2993 .min_rec_size = sizeof(struct bpf_func_info_min),
2994 .ext_info = &btf_ext->func_info,
2995 .desc = "func_info"
2996 };
2997
2998 return btf_ext_setup_info(btf_ext, ¶m);
2999 }
3000
btf_ext_setup_line_info(struct btf_ext * btf_ext)3001 static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
3002 {
3003 struct btf_ext_sec_setup_param param = {
3004 .off = btf_ext->hdr->line_info_off,
3005 .len = btf_ext->hdr->line_info_len,
3006 .min_rec_size = sizeof(struct bpf_line_info_min),
3007 .ext_info = &btf_ext->line_info,
3008 .desc = "line_info",
3009 };
3010
3011 return btf_ext_setup_info(btf_ext, ¶m);
3012 }
3013
btf_ext_setup_core_relos(struct btf_ext * btf_ext)3014 static int btf_ext_setup_core_relos(struct btf_ext *btf_ext)
3015 {
3016 struct btf_ext_sec_setup_param param = {
3017 .off = btf_ext->hdr->core_relo_off,
3018 .len = btf_ext->hdr->core_relo_len,
3019 .min_rec_size = sizeof(struct bpf_core_relo),
3020 .ext_info = &btf_ext->core_relo_info,
3021 .desc = "core_relo",
3022 };
3023
3024 return btf_ext_setup_info(btf_ext, ¶m);
3025 }
3026
btf_ext_parse_hdr(__u8 * data,__u32 data_size)3027 static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
3028 {
3029 const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
3030
3031 if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
3032 data_size < hdr->hdr_len) {
3033 pr_debug("BTF.ext header not found");
3034 return -EINVAL;
3035 }
3036
3037 if (hdr->magic == bswap_16(BTF_MAGIC)) {
3038 pr_warn("BTF.ext in non-native endianness is not supported\n");
3039 return -ENOTSUP;
3040 } else if (hdr->magic != BTF_MAGIC) {
3041 pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
3042 return -EINVAL;
3043 }
3044
3045 if (hdr->version != BTF_VERSION) {
3046 pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
3047 return -ENOTSUP;
3048 }
3049
3050 if (hdr->flags) {
3051 pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
3052 return -ENOTSUP;
3053 }
3054
3055 if (data_size == hdr->hdr_len) {
3056 pr_debug("BTF.ext has no data\n");
3057 return -EINVAL;
3058 }
3059
3060 return 0;
3061 }
3062
btf_ext__free(struct btf_ext * btf_ext)3063 void btf_ext__free(struct btf_ext *btf_ext)
3064 {
3065 if (IS_ERR_OR_NULL(btf_ext))
3066 return;
3067 free(btf_ext->func_info.sec_idxs);
3068 free(btf_ext->line_info.sec_idxs);
3069 free(btf_ext->core_relo_info.sec_idxs);
3070 free(btf_ext->data);
3071 free(btf_ext);
3072 }
3073
btf_ext__new(const __u8 * data,__u32 size)3074 struct btf_ext *btf_ext__new(const __u8 *data, __u32 size)
3075 {
3076 struct btf_ext *btf_ext;
3077 int err;
3078
3079 btf_ext = calloc(1, sizeof(struct btf_ext));
3080 if (!btf_ext)
3081 return libbpf_err_ptr(-ENOMEM);
3082
3083 btf_ext->data_size = size;
3084 btf_ext->data = malloc(size);
3085 if (!btf_ext->data) {
3086 err = -ENOMEM;
3087 goto done;
3088 }
3089 memcpy(btf_ext->data, data, size);
3090
3091 err = btf_ext_parse_hdr(btf_ext->data, size);
3092 if (err)
3093 goto done;
3094
3095 if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, line_info_len)) {
3096 err = -EINVAL;
3097 goto done;
3098 }
3099
3100 err = btf_ext_setup_func_info(btf_ext);
3101 if (err)
3102 goto done;
3103
3104 err = btf_ext_setup_line_info(btf_ext);
3105 if (err)
3106 goto done;
3107
3108 if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
3109 goto done; /* skip core relos parsing */
3110
3111 err = btf_ext_setup_core_relos(btf_ext);
3112 if (err)
3113 goto done;
3114
3115 done:
3116 if (err) {
3117 btf_ext__free(btf_ext);
3118 return libbpf_err_ptr(err);
3119 }
3120
3121 return btf_ext;
3122 }
3123
btf_ext__raw_data(const struct btf_ext * btf_ext,__u32 * size)3124 const void *btf_ext__raw_data(const struct btf_ext *btf_ext, __u32 *size)
3125 {
3126 *size = btf_ext->data_size;
3127 return btf_ext->data;
3128 }
3129
3130 __attribute__((alias("btf_ext__raw_data")))
3131 const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size);
3132
3133
3134 struct btf_dedup;
3135
3136 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts);
3137 static void btf_dedup_free(struct btf_dedup *d);
3138 static int btf_dedup_prep(struct btf_dedup *d);
3139 static int btf_dedup_strings(struct btf_dedup *d);
3140 static int btf_dedup_prim_types(struct btf_dedup *d);
3141 static int btf_dedup_struct_types(struct btf_dedup *d);
3142 static int btf_dedup_ref_types(struct btf_dedup *d);
3143 static int btf_dedup_resolve_fwds(struct btf_dedup *d);
3144 static int btf_dedup_compact_types(struct btf_dedup *d);
3145 static int btf_dedup_remap_types(struct btf_dedup *d);
3146
3147 /*
3148 * Deduplicate BTF types and strings.
3149 *
3150 * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
3151 * section with all BTF type descriptors and string data. It overwrites that
3152 * memory in-place with deduplicated types and strings without any loss of
3153 * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
3154 * is provided, all the strings referenced from .BTF.ext section are honored
3155 * and updated to point to the right offsets after deduplication.
3156 *
3157 * If function returns with error, type/string data might be garbled and should
3158 * be discarded.
3159 *
3160 * More verbose and detailed description of both problem btf_dedup is solving,
3161 * as well as solution could be found at:
3162 * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
3163 *
3164 * Problem description and justification
3165 * =====================================
3166 *
3167 * BTF type information is typically emitted either as a result of conversion
3168 * from DWARF to BTF or directly by compiler. In both cases, each compilation
3169 * unit contains information about a subset of all the types that are used
3170 * in an application. These subsets are frequently overlapping and contain a lot
3171 * of duplicated information when later concatenated together into a single
3172 * binary. This algorithm ensures that each unique type is represented by single
3173 * BTF type descriptor, greatly reducing resulting size of BTF data.
3174 *
3175 * Compilation unit isolation and subsequent duplication of data is not the only
3176 * problem. The same type hierarchy (e.g., struct and all the type that struct
3177 * references) in different compilation units can be represented in BTF to
3178 * various degrees of completeness (or, rather, incompleteness) due to
3179 * struct/union forward declarations.
3180 *
3181 * Let's take a look at an example, that we'll use to better understand the
3182 * problem (and solution). Suppose we have two compilation units, each using
3183 * same `struct S`, but each of them having incomplete type information about
3184 * struct's fields:
3185 *
3186 * // CU #1:
3187 * struct S;
3188 * struct A {
3189 * int a;
3190 * struct A* self;
3191 * struct S* parent;
3192 * };
3193 * struct B;
3194 * struct S {
3195 * struct A* a_ptr;
3196 * struct B* b_ptr;
3197 * };
3198 *
3199 * // CU #2:
3200 * struct S;
3201 * struct A;
3202 * struct B {
3203 * int b;
3204 * struct B* self;
3205 * struct S* parent;
3206 * };
3207 * struct S {
3208 * struct A* a_ptr;
3209 * struct B* b_ptr;
3210 * };
3211 *
3212 * In case of CU #1, BTF data will know only that `struct B` exist (but no
3213 * more), but will know the complete type information about `struct A`. While
3214 * for CU #2, it will know full type information about `struct B`, but will
3215 * only know about forward declaration of `struct A` (in BTF terms, it will
3216 * have `BTF_KIND_FWD` type descriptor with name `B`).
3217 *
3218 * This compilation unit isolation means that it's possible that there is no
3219 * single CU with complete type information describing structs `S`, `A`, and
3220 * `B`. Also, we might get tons of duplicated and redundant type information.
3221 *
3222 * Additional complication we need to keep in mind comes from the fact that
3223 * types, in general, can form graphs containing cycles, not just DAGs.
3224 *
3225 * While algorithm does deduplication, it also merges and resolves type
3226 * information (unless disabled throught `struct btf_opts`), whenever possible.
3227 * E.g., in the example above with two compilation units having partial type
3228 * information for structs `A` and `B`, the output of algorithm will emit
3229 * a single copy of each BTF type that describes structs `A`, `B`, and `S`
3230 * (as well as type information for `int` and pointers), as if they were defined
3231 * in a single compilation unit as:
3232 *
3233 * struct A {
3234 * int a;
3235 * struct A* self;
3236 * struct S* parent;
3237 * };
3238 * struct B {
3239 * int b;
3240 * struct B* self;
3241 * struct S* parent;
3242 * };
3243 * struct S {
3244 * struct A* a_ptr;
3245 * struct B* b_ptr;
3246 * };
3247 *
3248 * Algorithm summary
3249 * =================
3250 *
3251 * Algorithm completes its work in 7 separate passes:
3252 *
3253 * 1. Strings deduplication.
3254 * 2. Primitive types deduplication (int, enum, fwd).
3255 * 3. Struct/union types deduplication.
3256 * 4. Resolve unambiguous forward declarations.
3257 * 5. Reference types deduplication (pointers, typedefs, arrays, funcs, func
3258 * protos, and const/volatile/restrict modifiers).
3259 * 6. Types compaction.
3260 * 7. Types remapping.
3261 *
3262 * Algorithm determines canonical type descriptor, which is a single
3263 * representative type for each truly unique type. This canonical type is the
3264 * one that will go into final deduplicated BTF type information. For
3265 * struct/unions, it is also the type that algorithm will merge additional type
3266 * information into (while resolving FWDs), as it discovers it from data in
3267 * other CUs. Each input BTF type eventually gets either mapped to itself, if
3268 * that type is canonical, or to some other type, if that type is equivalent
3269 * and was chosen as canonical representative. This mapping is stored in
3270 * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
3271 * FWD type got resolved to.
3272 *
3273 * To facilitate fast discovery of canonical types, we also maintain canonical
3274 * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
3275 * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
3276 * that match that signature. With sufficiently good choice of type signature
3277 * hashing function, we can limit number of canonical types for each unique type
3278 * signature to a very small number, allowing to find canonical type for any
3279 * duplicated type very quickly.
3280 *
3281 * Struct/union deduplication is the most critical part and algorithm for
3282 * deduplicating structs/unions is described in greater details in comments for
3283 * `btf_dedup_is_equiv` function.
3284 */
btf__dedup(struct btf * btf,const struct btf_dedup_opts * opts)3285 int btf__dedup(struct btf *btf, const struct btf_dedup_opts *opts)
3286 {
3287 struct btf_dedup *d;
3288 int err;
3289
3290 if (!OPTS_VALID(opts, btf_dedup_opts))
3291 return libbpf_err(-EINVAL);
3292
3293 d = btf_dedup_new(btf, opts);
3294 if (IS_ERR(d)) {
3295 pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
3296 return libbpf_err(-EINVAL);
3297 }
3298
3299 if (btf_ensure_modifiable(btf)) {
3300 err = -ENOMEM;
3301 goto done;
3302 }
3303
3304 err = btf_dedup_prep(d);
3305 if (err) {
3306 pr_debug("btf_dedup_prep failed:%d\n", err);
3307 goto done;
3308 }
3309 err = btf_dedup_strings(d);
3310 if (err < 0) {
3311 pr_debug("btf_dedup_strings failed:%d\n", err);
3312 goto done;
3313 }
3314 err = btf_dedup_prim_types(d);
3315 if (err < 0) {
3316 pr_debug("btf_dedup_prim_types failed:%d\n", err);
3317 goto done;
3318 }
3319 err = btf_dedup_struct_types(d);
3320 if (err < 0) {
3321 pr_debug("btf_dedup_struct_types failed:%d\n", err);
3322 goto done;
3323 }
3324 err = btf_dedup_resolve_fwds(d);
3325 if (err < 0) {
3326 pr_debug("btf_dedup_resolve_fwds failed:%d\n", err);
3327 goto done;
3328 }
3329 err = btf_dedup_ref_types(d);
3330 if (err < 0) {
3331 pr_debug("btf_dedup_ref_types failed:%d\n", err);
3332 goto done;
3333 }
3334 err = btf_dedup_compact_types(d);
3335 if (err < 0) {
3336 pr_debug("btf_dedup_compact_types failed:%d\n", err);
3337 goto done;
3338 }
3339 err = btf_dedup_remap_types(d);
3340 if (err < 0) {
3341 pr_debug("btf_dedup_remap_types failed:%d\n", err);
3342 goto done;
3343 }
3344
3345 done:
3346 btf_dedup_free(d);
3347 return libbpf_err(err);
3348 }
3349
3350 #define BTF_UNPROCESSED_ID ((__u32)-1)
3351 #define BTF_IN_PROGRESS_ID ((__u32)-2)
3352
3353 struct btf_dedup {
3354 /* .BTF section to be deduped in-place */
3355 struct btf *btf;
3356 /*
3357 * Optional .BTF.ext section. When provided, any strings referenced
3358 * from it will be taken into account when deduping strings
3359 */
3360 struct btf_ext *btf_ext;
3361 /*
3362 * This is a map from any type's signature hash to a list of possible
3363 * canonical representative type candidates. Hash collisions are
3364 * ignored, so even types of various kinds can share same list of
3365 * candidates, which is fine because we rely on subsequent
3366 * btf_xxx_equal() checks to authoritatively verify type equality.
3367 */
3368 struct hashmap *dedup_table;
3369 /* Canonical types map */
3370 __u32 *map;
3371 /* Hypothetical mapping, used during type graph equivalence checks */
3372 __u32 *hypot_map;
3373 __u32 *hypot_list;
3374 size_t hypot_cnt;
3375 size_t hypot_cap;
3376 /* Whether hypothetical mapping, if successful, would need to adjust
3377 * already canonicalized types (due to a new forward declaration to
3378 * concrete type resolution). In such case, during split BTF dedup
3379 * candidate type would still be considered as different, because base
3380 * BTF is considered to be immutable.
3381 */
3382 bool hypot_adjust_canon;
3383 /* Various option modifying behavior of algorithm */
3384 struct btf_dedup_opts opts;
3385 /* temporary strings deduplication state */
3386 struct strset *strs_set;
3387 };
3388
hash_combine(long h,long value)3389 static long hash_combine(long h, long value)
3390 {
3391 return h * 31 + value;
3392 }
3393
3394 #define for_each_dedup_cand(d, node, hash) \
3395 hashmap__for_each_key_entry(d->dedup_table, node, hash)
3396
btf_dedup_table_add(struct btf_dedup * d,long hash,__u32 type_id)3397 static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
3398 {
3399 return hashmap__append(d->dedup_table, hash, type_id);
3400 }
3401
btf_dedup_hypot_map_add(struct btf_dedup * d,__u32 from_id,__u32 to_id)3402 static int btf_dedup_hypot_map_add(struct btf_dedup *d,
3403 __u32 from_id, __u32 to_id)
3404 {
3405 if (d->hypot_cnt == d->hypot_cap) {
3406 __u32 *new_list;
3407
3408 d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
3409 new_list = libbpf_reallocarray(d->hypot_list, d->hypot_cap, sizeof(__u32));
3410 if (!new_list)
3411 return -ENOMEM;
3412 d->hypot_list = new_list;
3413 }
3414 d->hypot_list[d->hypot_cnt++] = from_id;
3415 d->hypot_map[from_id] = to_id;
3416 return 0;
3417 }
3418
btf_dedup_clear_hypot_map(struct btf_dedup * d)3419 static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
3420 {
3421 int i;
3422
3423 for (i = 0; i < d->hypot_cnt; i++)
3424 d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
3425 d->hypot_cnt = 0;
3426 d->hypot_adjust_canon = false;
3427 }
3428
btf_dedup_free(struct btf_dedup * d)3429 static void btf_dedup_free(struct btf_dedup *d)
3430 {
3431 hashmap__free(d->dedup_table);
3432 d->dedup_table = NULL;
3433
3434 free(d->map);
3435 d->map = NULL;
3436
3437 free(d->hypot_map);
3438 d->hypot_map = NULL;
3439
3440 free(d->hypot_list);
3441 d->hypot_list = NULL;
3442
3443 free(d);
3444 }
3445
btf_dedup_identity_hash_fn(long key,void * ctx)3446 static size_t btf_dedup_identity_hash_fn(long key, void *ctx)
3447 {
3448 return key;
3449 }
3450
btf_dedup_collision_hash_fn(long key,void * ctx)3451 static size_t btf_dedup_collision_hash_fn(long key, void *ctx)
3452 {
3453 return 0;
3454 }
3455
btf_dedup_equal_fn(long k1,long k2,void * ctx)3456 static bool btf_dedup_equal_fn(long k1, long k2, void *ctx)
3457 {
3458 return k1 == k2;
3459 }
3460
btf_dedup_new(struct btf * btf,const struct btf_dedup_opts * opts)3461 static struct btf_dedup *btf_dedup_new(struct btf *btf, const struct btf_dedup_opts *opts)
3462 {
3463 struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
3464 hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
3465 int i, err = 0, type_cnt;
3466
3467 if (!d)
3468 return ERR_PTR(-ENOMEM);
3469
3470 if (OPTS_GET(opts, force_collisions, false))
3471 hash_fn = btf_dedup_collision_hash_fn;
3472
3473 d->btf = btf;
3474 d->btf_ext = OPTS_GET(opts, btf_ext, NULL);
3475
3476 d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
3477 if (IS_ERR(d->dedup_table)) {
3478 err = PTR_ERR(d->dedup_table);
3479 d->dedup_table = NULL;
3480 goto done;
3481 }
3482
3483 type_cnt = btf__type_cnt(btf);
3484 d->map = malloc(sizeof(__u32) * type_cnt);
3485 if (!d->map) {
3486 err = -ENOMEM;
3487 goto done;
3488 }
3489 /* special BTF "void" type is made canonical immediately */
3490 d->map[0] = 0;
3491 for (i = 1; i < type_cnt; i++) {
3492 struct btf_type *t = btf_type_by_id(d->btf, i);
3493
3494 /* VAR and DATASEC are never deduped and are self-canonical */
3495 if (btf_is_var(t) || btf_is_datasec(t))
3496 d->map[i] = i;
3497 else
3498 d->map[i] = BTF_UNPROCESSED_ID;
3499 }
3500
3501 d->hypot_map = malloc(sizeof(__u32) * type_cnt);
3502 if (!d->hypot_map) {
3503 err = -ENOMEM;
3504 goto done;
3505 }
3506 for (i = 0; i < type_cnt; i++)
3507 d->hypot_map[i] = BTF_UNPROCESSED_ID;
3508
3509 done:
3510 if (err) {
3511 btf_dedup_free(d);
3512 return ERR_PTR(err);
3513 }
3514
3515 return d;
3516 }
3517
3518 /*
3519 * Iterate over all possible places in .BTF and .BTF.ext that can reference
3520 * string and pass pointer to it to a provided callback `fn`.
3521 */
btf_for_each_str_off(struct btf_dedup * d,str_off_visit_fn fn,void * ctx)3522 static int btf_for_each_str_off(struct btf_dedup *d, str_off_visit_fn fn, void *ctx)
3523 {
3524 int i, r;
3525
3526 for (i = 0; i < d->btf->nr_types; i++) {
3527 struct btf_field_iter it;
3528 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
3529 __u32 *str_off;
3530
3531 r = btf_field_iter_init(&it, t, BTF_FIELD_ITER_STRS);
3532 if (r)
3533 return r;
3534
3535 while ((str_off = btf_field_iter_next(&it))) {
3536 r = fn(str_off, ctx);
3537 if (r)
3538 return r;
3539 }
3540 }
3541
3542 if (!d->btf_ext)
3543 return 0;
3544
3545 r = btf_ext_visit_str_offs(d->btf_ext, fn, ctx);
3546 if (r)
3547 return r;
3548
3549 return 0;
3550 }
3551
strs_dedup_remap_str_off(__u32 * str_off_ptr,void * ctx)3552 static int strs_dedup_remap_str_off(__u32 *str_off_ptr, void *ctx)
3553 {
3554 struct btf_dedup *d = ctx;
3555 __u32 str_off = *str_off_ptr;
3556 const char *s;
3557 int off, err;
3558
3559 /* don't touch empty string or string in main BTF */
3560 if (str_off == 0 || str_off < d->btf->start_str_off)
3561 return 0;
3562
3563 s = btf__str_by_offset(d->btf, str_off);
3564 if (d->btf->base_btf) {
3565 err = btf__find_str(d->btf->base_btf, s);
3566 if (err >= 0) {
3567 *str_off_ptr = err;
3568 return 0;
3569 }
3570 if (err != -ENOENT)
3571 return err;
3572 }
3573
3574 off = strset__add_str(d->strs_set, s);
3575 if (off < 0)
3576 return off;
3577
3578 *str_off_ptr = d->btf->start_str_off + off;
3579 return 0;
3580 }
3581
3582 /*
3583 * Dedup string and filter out those that are not referenced from either .BTF
3584 * or .BTF.ext (if provided) sections.
3585 *
3586 * This is done by building index of all strings in BTF's string section,
3587 * then iterating over all entities that can reference strings (e.g., type
3588 * names, struct field names, .BTF.ext line info, etc) and marking corresponding
3589 * strings as used. After that all used strings are deduped and compacted into
3590 * sequential blob of memory and new offsets are calculated. Then all the string
3591 * references are iterated again and rewritten using new offsets.
3592 */
btf_dedup_strings(struct btf_dedup * d)3593 static int btf_dedup_strings(struct btf_dedup *d)
3594 {
3595 int err;
3596
3597 if (d->btf->strs_deduped)
3598 return 0;
3599
3600 d->strs_set = strset__new(BTF_MAX_STR_OFFSET, NULL, 0);
3601 if (IS_ERR(d->strs_set)) {
3602 err = PTR_ERR(d->strs_set);
3603 goto err_out;
3604 }
3605
3606 if (!d->btf->base_btf) {
3607 /* insert empty string; we won't be looking it up during strings
3608 * dedup, but it's good to have it for generic BTF string lookups
3609 */
3610 err = strset__add_str(d->strs_set, "");
3611 if (err < 0)
3612 goto err_out;
3613 }
3614
3615 /* remap string offsets */
3616 err = btf_for_each_str_off(d, strs_dedup_remap_str_off, d);
3617 if (err)
3618 goto err_out;
3619
3620 /* replace BTF string data and hash with deduped ones */
3621 strset__free(d->btf->strs_set);
3622 d->btf->hdr->str_len = strset__data_size(d->strs_set);
3623 d->btf->strs_set = d->strs_set;
3624 d->strs_set = NULL;
3625 d->btf->strs_deduped = true;
3626 return 0;
3627
3628 err_out:
3629 strset__free(d->strs_set);
3630 d->strs_set = NULL;
3631
3632 return err;
3633 }
3634
btf_hash_common(struct btf_type * t)3635 static long btf_hash_common(struct btf_type *t)
3636 {
3637 long h;
3638
3639 h = hash_combine(0, t->name_off);
3640 h = hash_combine(h, t->info);
3641 h = hash_combine(h, t->size);
3642 return h;
3643 }
3644
btf_equal_common(struct btf_type * t1,struct btf_type * t2)3645 static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
3646 {
3647 return t1->name_off == t2->name_off &&
3648 t1->info == t2->info &&
3649 t1->size == t2->size;
3650 }
3651
3652 /* Calculate type signature hash of INT or TAG. */
btf_hash_int_decl_tag(struct btf_type * t)3653 static long btf_hash_int_decl_tag(struct btf_type *t)
3654 {
3655 __u32 info = *(__u32 *)(t + 1);
3656 long h;
3657
3658 h = btf_hash_common(t);
3659 h = hash_combine(h, info);
3660 return h;
3661 }
3662
3663 /* Check structural equality of two INTs or TAGs. */
btf_equal_int_tag(struct btf_type * t1,struct btf_type * t2)3664 static bool btf_equal_int_tag(struct btf_type *t1, struct btf_type *t2)
3665 {
3666 __u32 info1, info2;
3667
3668 if (!btf_equal_common(t1, t2))
3669 return false;
3670 info1 = *(__u32 *)(t1 + 1);
3671 info2 = *(__u32 *)(t2 + 1);
3672 return info1 == info2;
3673 }
3674
3675 /* Calculate type signature hash of ENUM/ENUM64. */
btf_hash_enum(struct btf_type * t)3676 static long btf_hash_enum(struct btf_type *t)
3677 {
3678 long h;
3679
3680 /* don't hash vlen, enum members and size to support enum fwd resolving */
3681 h = hash_combine(0, t->name_off);
3682 return h;
3683 }
3684
btf_equal_enum_members(struct btf_type * t1,struct btf_type * t2)3685 static bool btf_equal_enum_members(struct btf_type *t1, struct btf_type *t2)
3686 {
3687 const struct btf_enum *m1, *m2;
3688 __u16 vlen;
3689 int i;
3690
3691 vlen = btf_vlen(t1);
3692 m1 = btf_enum(t1);
3693 m2 = btf_enum(t2);
3694 for (i = 0; i < vlen; i++) {
3695 if (m1->name_off != m2->name_off || m1->val != m2->val)
3696 return false;
3697 m1++;
3698 m2++;
3699 }
3700 return true;
3701 }
3702
btf_equal_enum64_members(struct btf_type * t1,struct btf_type * t2)3703 static bool btf_equal_enum64_members(struct btf_type *t1, struct btf_type *t2)
3704 {
3705 const struct btf_enum64 *m1, *m2;
3706 __u16 vlen;
3707 int i;
3708
3709 vlen = btf_vlen(t1);
3710 m1 = btf_enum64(t1);
3711 m2 = btf_enum64(t2);
3712 for (i = 0; i < vlen; i++) {
3713 if (m1->name_off != m2->name_off || m1->val_lo32 != m2->val_lo32 ||
3714 m1->val_hi32 != m2->val_hi32)
3715 return false;
3716 m1++;
3717 m2++;
3718 }
3719 return true;
3720 }
3721
3722 /* Check structural equality of two ENUMs or ENUM64s. */
btf_equal_enum(struct btf_type * t1,struct btf_type * t2)3723 static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
3724 {
3725 if (!btf_equal_common(t1, t2))
3726 return false;
3727
3728 /* t1 & t2 kinds are identical because of btf_equal_common */
3729 if (btf_kind(t1) == BTF_KIND_ENUM)
3730 return btf_equal_enum_members(t1, t2);
3731 else
3732 return btf_equal_enum64_members(t1, t2);
3733 }
3734
btf_is_enum_fwd(struct btf_type * t)3735 static inline bool btf_is_enum_fwd(struct btf_type *t)
3736 {
3737 return btf_is_any_enum(t) && btf_vlen(t) == 0;
3738 }
3739
btf_compat_enum(struct btf_type * t1,struct btf_type * t2)3740 static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
3741 {
3742 if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3743 return btf_equal_enum(t1, t2);
3744 /* At this point either t1 or t2 or both are forward declarations, thus:
3745 * - skip comparing vlen because it is zero for forward declarations;
3746 * - skip comparing size to allow enum forward declarations
3747 * to be compatible with enum64 full declarations;
3748 * - skip comparing kind for the same reason.
3749 */
3750 return t1->name_off == t2->name_off &&
3751 btf_is_any_enum(t1) && btf_is_any_enum(t2);
3752 }
3753
3754 /*
3755 * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
3756 * as referenced type IDs equivalence is established separately during type
3757 * graph equivalence check algorithm.
3758 */
btf_hash_struct(struct btf_type * t)3759 static long btf_hash_struct(struct btf_type *t)
3760 {
3761 const struct btf_member *member = btf_members(t);
3762 __u32 vlen = btf_vlen(t);
3763 long h = btf_hash_common(t);
3764 int i;
3765
3766 for (i = 0; i < vlen; i++) {
3767 h = hash_combine(h, member->name_off);
3768 h = hash_combine(h, member->offset);
3769 /* no hashing of referenced type ID, it can be unresolved yet */
3770 member++;
3771 }
3772 return h;
3773 }
3774
3775 /*
3776 * Check structural compatibility of two STRUCTs/UNIONs, ignoring referenced
3777 * type IDs. This check is performed during type graph equivalence check and
3778 * referenced types equivalence is checked separately.
3779 */
btf_shallow_equal_struct(struct btf_type * t1,struct btf_type * t2)3780 static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
3781 {
3782 const struct btf_member *m1, *m2;
3783 __u16 vlen;
3784 int i;
3785
3786 if (!btf_equal_common(t1, t2))
3787 return false;
3788
3789 vlen = btf_vlen(t1);
3790 m1 = btf_members(t1);
3791 m2 = btf_members(t2);
3792 for (i = 0; i < vlen; i++) {
3793 if (m1->name_off != m2->name_off || m1->offset != m2->offset)
3794 return false;
3795 m1++;
3796 m2++;
3797 }
3798 return true;
3799 }
3800
3801 /*
3802 * Calculate type signature hash of ARRAY, including referenced type IDs,
3803 * under assumption that they were already resolved to canonical type IDs and
3804 * are not going to change.
3805 */
btf_hash_array(struct btf_type * t)3806 static long btf_hash_array(struct btf_type *t)
3807 {
3808 const struct btf_array *info = btf_array(t);
3809 long h = btf_hash_common(t);
3810
3811 h = hash_combine(h, info->type);
3812 h = hash_combine(h, info->index_type);
3813 h = hash_combine(h, info->nelems);
3814 return h;
3815 }
3816
3817 /*
3818 * Check exact equality of two ARRAYs, taking into account referenced
3819 * type IDs, under assumption that they were already resolved to canonical
3820 * type IDs and are not going to change.
3821 * This function is called during reference types deduplication to compare
3822 * ARRAY to potential canonical representative.
3823 */
btf_equal_array(struct btf_type * t1,struct btf_type * t2)3824 static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
3825 {
3826 const struct btf_array *info1, *info2;
3827
3828 if (!btf_equal_common(t1, t2))
3829 return false;
3830
3831 info1 = btf_array(t1);
3832 info2 = btf_array(t2);
3833 return info1->type == info2->type &&
3834 info1->index_type == info2->index_type &&
3835 info1->nelems == info2->nelems;
3836 }
3837
3838 /*
3839 * Check structural compatibility of two ARRAYs, ignoring referenced type
3840 * IDs. This check is performed during type graph equivalence check and
3841 * referenced types equivalence is checked separately.
3842 */
btf_compat_array(struct btf_type * t1,struct btf_type * t2)3843 static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
3844 {
3845 if (!btf_equal_common(t1, t2))
3846 return false;
3847
3848 return btf_array(t1)->nelems == btf_array(t2)->nelems;
3849 }
3850
3851 /*
3852 * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
3853 * under assumption that they were already resolved to canonical type IDs and
3854 * are not going to change.
3855 */
btf_hash_fnproto(struct btf_type * t)3856 static long btf_hash_fnproto(struct btf_type *t)
3857 {
3858 const struct btf_param *member = btf_params(t);
3859 __u16 vlen = btf_vlen(t);
3860 long h = btf_hash_common(t);
3861 int i;
3862
3863 for (i = 0; i < vlen; i++) {
3864 h = hash_combine(h, member->name_off);
3865 h = hash_combine(h, member->type);
3866 member++;
3867 }
3868 return h;
3869 }
3870
3871 /*
3872 * Check exact equality of two FUNC_PROTOs, taking into account referenced
3873 * type IDs, under assumption that they were already resolved to canonical
3874 * type IDs and are not going to change.
3875 * This function is called during reference types deduplication to compare
3876 * FUNC_PROTO to potential canonical representative.
3877 */
btf_equal_fnproto(struct btf_type * t1,struct btf_type * t2)3878 static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
3879 {
3880 const struct btf_param *m1, *m2;
3881 __u16 vlen;
3882 int i;
3883
3884 if (!btf_equal_common(t1, t2))
3885 return false;
3886
3887 vlen = btf_vlen(t1);
3888 m1 = btf_params(t1);
3889 m2 = btf_params(t2);
3890 for (i = 0; i < vlen; i++) {
3891 if (m1->name_off != m2->name_off || m1->type != m2->type)
3892 return false;
3893 m1++;
3894 m2++;
3895 }
3896 return true;
3897 }
3898
3899 /*
3900 * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3901 * IDs. This check is performed during type graph equivalence check and
3902 * referenced types equivalence is checked separately.
3903 */
btf_compat_fnproto(struct btf_type * t1,struct btf_type * t2)3904 static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
3905 {
3906 const struct btf_param *m1, *m2;
3907 __u16 vlen;
3908 int i;
3909
3910 /* skip return type ID */
3911 if (t1->name_off != t2->name_off || t1->info != t2->info)
3912 return false;
3913
3914 vlen = btf_vlen(t1);
3915 m1 = btf_params(t1);
3916 m2 = btf_params(t2);
3917 for (i = 0; i < vlen; i++) {
3918 if (m1->name_off != m2->name_off)
3919 return false;
3920 m1++;
3921 m2++;
3922 }
3923 return true;
3924 }
3925
3926 /* Prepare split BTF for deduplication by calculating hashes of base BTF's
3927 * types and initializing the rest of the state (canonical type mapping) for
3928 * the fixed base BTF part.
3929 */
btf_dedup_prep(struct btf_dedup * d)3930 static int btf_dedup_prep(struct btf_dedup *d)
3931 {
3932 struct btf_type *t;
3933 int type_id;
3934 long h;
3935
3936 if (!d->btf->base_btf)
3937 return 0;
3938
3939 for (type_id = 1; type_id < d->btf->start_id; type_id++) {
3940 t = btf_type_by_id(d->btf, type_id);
3941
3942 /* all base BTF types are self-canonical by definition */
3943 d->map[type_id] = type_id;
3944
3945 switch (btf_kind(t)) {
3946 case BTF_KIND_VAR:
3947 case BTF_KIND_DATASEC:
3948 /* VAR and DATASEC are never hash/deduplicated */
3949 continue;
3950 case BTF_KIND_CONST:
3951 case BTF_KIND_VOLATILE:
3952 case BTF_KIND_RESTRICT:
3953 case BTF_KIND_PTR:
3954 case BTF_KIND_FWD:
3955 case BTF_KIND_TYPEDEF:
3956 case BTF_KIND_FUNC:
3957 case BTF_KIND_FLOAT:
3958 case BTF_KIND_TYPE_TAG:
3959 h = btf_hash_common(t);
3960 break;
3961 case BTF_KIND_INT:
3962 case BTF_KIND_DECL_TAG:
3963 h = btf_hash_int_decl_tag(t);
3964 break;
3965 case BTF_KIND_ENUM:
3966 case BTF_KIND_ENUM64:
3967 h = btf_hash_enum(t);
3968 break;
3969 case BTF_KIND_STRUCT:
3970 case BTF_KIND_UNION:
3971 h = btf_hash_struct(t);
3972 break;
3973 case BTF_KIND_ARRAY:
3974 h = btf_hash_array(t);
3975 break;
3976 case BTF_KIND_FUNC_PROTO:
3977 h = btf_hash_fnproto(t);
3978 break;
3979 default:
3980 pr_debug("unknown kind %d for type [%d]\n", btf_kind(t), type_id);
3981 return -EINVAL;
3982 }
3983 if (btf_dedup_table_add(d, h, type_id))
3984 return -ENOMEM;
3985 }
3986
3987 return 0;
3988 }
3989
3990 /*
3991 * Deduplicate primitive types, that can't reference other types, by calculating
3992 * their type signature hash and comparing them with any possible canonical
3993 * candidate. If no canonical candidate matches, type itself is marked as
3994 * canonical and is added into `btf_dedup->dedup_table` as another candidate.
3995 */
btf_dedup_prim_type(struct btf_dedup * d,__u32 type_id)3996 static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
3997 {
3998 struct btf_type *t = btf_type_by_id(d->btf, type_id);
3999 struct hashmap_entry *hash_entry;
4000 struct btf_type *cand;
4001 /* if we don't find equivalent type, then we are canonical */
4002 __u32 new_id = type_id;
4003 __u32 cand_id;
4004 long h;
4005
4006 switch (btf_kind(t)) {
4007 case BTF_KIND_CONST:
4008 case BTF_KIND_VOLATILE:
4009 case BTF_KIND_RESTRICT:
4010 case BTF_KIND_PTR:
4011 case BTF_KIND_TYPEDEF:
4012 case BTF_KIND_ARRAY:
4013 case BTF_KIND_STRUCT:
4014 case BTF_KIND_UNION:
4015 case BTF_KIND_FUNC:
4016 case BTF_KIND_FUNC_PROTO:
4017 case BTF_KIND_VAR:
4018 case BTF_KIND_DATASEC:
4019 case BTF_KIND_DECL_TAG:
4020 case BTF_KIND_TYPE_TAG:
4021 return 0;
4022
4023 case BTF_KIND_INT:
4024 h = btf_hash_int_decl_tag(t);
4025 for_each_dedup_cand(d, hash_entry, h) {
4026 cand_id = hash_entry->value;
4027 cand = btf_type_by_id(d->btf, cand_id);
4028 if (btf_equal_int_tag(t, cand)) {
4029 new_id = cand_id;
4030 break;
4031 }
4032 }
4033 break;
4034
4035 case BTF_KIND_ENUM:
4036 case BTF_KIND_ENUM64:
4037 h = btf_hash_enum(t);
4038 for_each_dedup_cand(d, hash_entry, h) {
4039 cand_id = hash_entry->value;
4040 cand = btf_type_by_id(d->btf, cand_id);
4041 if (btf_equal_enum(t, cand)) {
4042 new_id = cand_id;
4043 break;
4044 }
4045 if (btf_compat_enum(t, cand)) {
4046 if (btf_is_enum_fwd(t)) {
4047 /* resolve fwd to full enum */
4048 new_id = cand_id;
4049 break;
4050 }
4051 /* resolve canonical enum fwd to full enum */
4052 d->map[cand_id] = type_id;
4053 }
4054 }
4055 break;
4056
4057 case BTF_KIND_FWD:
4058 case BTF_KIND_FLOAT:
4059 h = btf_hash_common(t);
4060 for_each_dedup_cand(d, hash_entry, h) {
4061 cand_id = hash_entry->value;
4062 cand = btf_type_by_id(d->btf, cand_id);
4063 if (btf_equal_common(t, cand)) {
4064 new_id = cand_id;
4065 break;
4066 }
4067 }
4068 break;
4069
4070 default:
4071 return -EINVAL;
4072 }
4073
4074 d->map[type_id] = new_id;
4075 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4076 return -ENOMEM;
4077
4078 return 0;
4079 }
4080
btf_dedup_prim_types(struct btf_dedup * d)4081 static int btf_dedup_prim_types(struct btf_dedup *d)
4082 {
4083 int i, err;
4084
4085 for (i = 0; i < d->btf->nr_types; i++) {
4086 err = btf_dedup_prim_type(d, d->btf->start_id + i);
4087 if (err)
4088 return err;
4089 }
4090 return 0;
4091 }
4092
4093 /*
4094 * Check whether type is already mapped into canonical one (could be to itself).
4095 */
is_type_mapped(struct btf_dedup * d,uint32_t type_id)4096 static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
4097 {
4098 return d->map[type_id] <= BTF_MAX_NR_TYPES;
4099 }
4100
4101 /*
4102 * Resolve type ID into its canonical type ID, if any; otherwise return original
4103 * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
4104 * STRUCT/UNION link and resolve it into canonical type ID as well.
4105 */
resolve_type_id(struct btf_dedup * d,__u32 type_id)4106 static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
4107 {
4108 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
4109 type_id = d->map[type_id];
4110 return type_id;
4111 }
4112
4113 /*
4114 * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
4115 * type ID.
4116 */
resolve_fwd_id(struct btf_dedup * d,uint32_t type_id)4117 static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
4118 {
4119 __u32 orig_type_id = type_id;
4120
4121 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
4122 return type_id;
4123
4124 while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
4125 type_id = d->map[type_id];
4126
4127 if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
4128 return type_id;
4129
4130 return orig_type_id;
4131 }
4132
4133
btf_fwd_kind(struct btf_type * t)4134 static inline __u16 btf_fwd_kind(struct btf_type *t)
4135 {
4136 return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
4137 }
4138
4139 /* Check if given two types are identical ARRAY definitions */
btf_dedup_identical_arrays(struct btf_dedup * d,__u32 id1,__u32 id2)4140 static bool btf_dedup_identical_arrays(struct btf_dedup *d, __u32 id1, __u32 id2)
4141 {
4142 struct btf_type *t1, *t2;
4143
4144 t1 = btf_type_by_id(d->btf, id1);
4145 t2 = btf_type_by_id(d->btf, id2);
4146 if (!btf_is_array(t1) || !btf_is_array(t2))
4147 return false;
4148
4149 return btf_equal_array(t1, t2);
4150 }
4151
4152 /* Check if given two types are identical STRUCT/UNION definitions */
btf_dedup_identical_structs(struct btf_dedup * d,__u32 id1,__u32 id2)4153 static bool btf_dedup_identical_structs(struct btf_dedup *d, __u32 id1, __u32 id2)
4154 {
4155 const struct btf_member *m1, *m2;
4156 struct btf_type *t1, *t2;
4157 int n, i;
4158
4159 t1 = btf_type_by_id(d->btf, id1);
4160 t2 = btf_type_by_id(d->btf, id2);
4161
4162 if (!btf_is_composite(t1) || btf_kind(t1) != btf_kind(t2))
4163 return false;
4164
4165 if (!btf_shallow_equal_struct(t1, t2))
4166 return false;
4167
4168 m1 = btf_members(t1);
4169 m2 = btf_members(t2);
4170 for (i = 0, n = btf_vlen(t1); i < n; i++, m1++, m2++) {
4171 if (m1->type != m2->type &&
4172 !btf_dedup_identical_arrays(d, m1->type, m2->type) &&
4173 !btf_dedup_identical_structs(d, m1->type, m2->type))
4174 return false;
4175 }
4176 return true;
4177 }
4178
btf_dedup_identical_ptrs(struct btf_dedup * d,__u32 id1,__u32 id2)4179 static bool btf_dedup_identical_ptrs(struct btf_dedup *d, __u32 id1, __u32 id2)
4180 {
4181 struct btf_type *t1, *t2;
4182
4183 t1 = btf_type_by_id(d->btf, id1);
4184 t2 = btf_type_by_id(d->btf, id2);
4185
4186 if (!btf_is_ptr(t1) || !btf_is_ptr(t2))
4187 return false;
4188
4189 return t1->type == t2->type;
4190 }
4191
4192 /*
4193 * Check equivalence of BTF type graph formed by candidate struct/union (we'll
4194 * call it "candidate graph" in this description for brevity) to a type graph
4195 * formed by (potential) canonical struct/union ("canonical graph" for brevity
4196 * here, though keep in mind that not all types in canonical graph are
4197 * necessarily canonical representatives themselves, some of them might be
4198 * duplicates or its uniqueness might not have been established yet).
4199 * Returns:
4200 * - >0, if type graphs are equivalent;
4201 * - 0, if not equivalent;
4202 * - <0, on error.
4203 *
4204 * Algorithm performs side-by-side DFS traversal of both type graphs and checks
4205 * equivalence of BTF types at each step. If at any point BTF types in candidate
4206 * and canonical graphs are not compatible structurally, whole graphs are
4207 * incompatible. If types are structurally equivalent (i.e., all information
4208 * except referenced type IDs is exactly the same), a mapping from `canon_id` to
4209 * a `cand_id` is recoded in hypothetical mapping (`btf_dedup->hypot_map`).
4210 * If a type references other types, then those referenced types are checked
4211 * for equivalence recursively.
4212 *
4213 * During DFS traversal, if we find that for current `canon_id` type we
4214 * already have some mapping in hypothetical map, we check for two possible
4215 * situations:
4216 * - `canon_id` is mapped to exactly the same type as `cand_id`. This will
4217 * happen when type graphs have cycles. In this case we assume those two
4218 * types are equivalent.
4219 * - `canon_id` is mapped to different type. This is contradiction in our
4220 * hypothetical mapping, because same graph in canonical graph corresponds
4221 * to two different types in candidate graph, which for equivalent type
4222 * graphs shouldn't happen. This condition terminates equivalence check
4223 * with negative result.
4224 *
4225 * If type graphs traversal exhausts types to check and find no contradiction,
4226 * then type graphs are equivalent.
4227 *
4228 * When checking types for equivalence, there is one special case: FWD types.
4229 * If FWD type resolution is allowed and one of the types (either from canonical
4230 * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
4231 * flag) and their names match, hypothetical mapping is updated to point from
4232 * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
4233 * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
4234 *
4235 * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
4236 * if there are two exactly named (or anonymous) structs/unions that are
4237 * compatible structurally, one of which has FWD field, while other is concrete
4238 * STRUCT/UNION, but according to C sources they are different structs/unions
4239 * that are referencing different types with the same name. This is extremely
4240 * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
4241 * this logic is causing problems.
4242 *
4243 * Doing FWD resolution means that both candidate and/or canonical graphs can
4244 * consists of portions of the graph that come from multiple compilation units.
4245 * This is due to the fact that types within single compilation unit are always
4246 * deduplicated and FWDs are already resolved, if referenced struct/union
4247 * definition is available. So, if we had unresolved FWD and found corresponding
4248 * STRUCT/UNION, they will be from different compilation units. This
4249 * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
4250 * type graph will likely have at least two different BTF types that describe
4251 * same type (e.g., most probably there will be two different BTF types for the
4252 * same 'int' primitive type) and could even have "overlapping" parts of type
4253 * graph that describe same subset of types.
4254 *
4255 * This in turn means that our assumption that each type in canonical graph
4256 * must correspond to exactly one type in candidate graph might not hold
4257 * anymore and will make it harder to detect contradictions using hypothetical
4258 * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
4259 * resolution only in canonical graph. FWDs in candidate graphs are never
4260 * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
4261 * that can occur:
4262 * - Both types in canonical and candidate graphs are FWDs. If they are
4263 * structurally equivalent, then they can either be both resolved to the
4264 * same STRUCT/UNION or not resolved at all. In both cases they are
4265 * equivalent and there is no need to resolve FWD on candidate side.
4266 * - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
4267 * so nothing to resolve as well, algorithm will check equivalence anyway.
4268 * - Type in canonical graph is FWD, while type in candidate is concrete
4269 * STRUCT/UNION. In this case candidate graph comes from single compilation
4270 * unit, so there is exactly one BTF type for each unique C type. After
4271 * resolving FWD into STRUCT/UNION, there might be more than one BTF type
4272 * in canonical graph mapping to single BTF type in candidate graph, but
4273 * because hypothetical mapping maps from canonical to candidate types, it's
4274 * alright, and we still maintain the property of having single `canon_id`
4275 * mapping to single `cand_id` (there could be two different `canon_id`
4276 * mapped to the same `cand_id`, but it's not contradictory).
4277 * - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
4278 * graph is FWD. In this case we are just going to check compatibility of
4279 * STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
4280 * assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
4281 * a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
4282 * turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
4283 * canonical graph.
4284 */
btf_dedup_is_equiv(struct btf_dedup * d,__u32 cand_id,__u32 canon_id)4285 static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
4286 __u32 canon_id)
4287 {
4288 struct btf_type *cand_type;
4289 struct btf_type *canon_type;
4290 __u32 hypot_type_id;
4291 __u16 cand_kind;
4292 __u16 canon_kind;
4293 int i, eq;
4294
4295 /* if both resolve to the same canonical, they must be equivalent */
4296 if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
4297 return 1;
4298
4299 canon_id = resolve_fwd_id(d, canon_id);
4300
4301 hypot_type_id = d->hypot_map[canon_id];
4302 if (hypot_type_id <= BTF_MAX_NR_TYPES) {
4303 if (hypot_type_id == cand_id)
4304 return 1;
4305 /* In some cases compiler will generate different DWARF types
4306 * for *identical* array type definitions and use them for
4307 * different fields within the *same* struct. This breaks type
4308 * equivalence check, which makes an assumption that candidate
4309 * types sub-graph has a consistent and deduped-by-compiler
4310 * types within a single CU. So work around that by explicitly
4311 * allowing identical array types here.
4312 */
4313 if (btf_dedup_identical_arrays(d, hypot_type_id, cand_id))
4314 return 1;
4315 /* It turns out that similar situation can happen with
4316 * struct/union sometimes, sigh... Handle the case where
4317 * structs/unions are exactly the same, down to the referenced
4318 * type IDs. Anything more complicated (e.g., if referenced
4319 * types are different, but equivalent) is *way more*
4320 * complicated and requires a many-to-many equivalence mapping.
4321 */
4322 if (btf_dedup_identical_structs(d, hypot_type_id, cand_id))
4323 return 1;
4324 /* A similar case is again observed for PTRs. */
4325 if (btf_dedup_identical_ptrs(d, hypot_type_id, cand_id))
4326 return 1;
4327 return 0;
4328 }
4329
4330 if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
4331 return -ENOMEM;
4332
4333 cand_type = btf_type_by_id(d->btf, cand_id);
4334 canon_type = btf_type_by_id(d->btf, canon_id);
4335 cand_kind = btf_kind(cand_type);
4336 canon_kind = btf_kind(canon_type);
4337
4338 if (cand_type->name_off != canon_type->name_off)
4339 return 0;
4340
4341 /* FWD <--> STRUCT/UNION equivalence check, if enabled */
4342 if ((cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
4343 && cand_kind != canon_kind) {
4344 __u16 real_kind;
4345 __u16 fwd_kind;
4346
4347 if (cand_kind == BTF_KIND_FWD) {
4348 real_kind = canon_kind;
4349 fwd_kind = btf_fwd_kind(cand_type);
4350 } else {
4351 real_kind = cand_kind;
4352 fwd_kind = btf_fwd_kind(canon_type);
4353 /* we'd need to resolve base FWD to STRUCT/UNION */
4354 if (fwd_kind == real_kind && canon_id < d->btf->start_id)
4355 d->hypot_adjust_canon = true;
4356 }
4357 return fwd_kind == real_kind;
4358 }
4359
4360 if (cand_kind != canon_kind)
4361 return 0;
4362
4363 switch (cand_kind) {
4364 case BTF_KIND_INT:
4365 return btf_equal_int_tag(cand_type, canon_type);
4366
4367 case BTF_KIND_ENUM:
4368 case BTF_KIND_ENUM64:
4369 return btf_compat_enum(cand_type, canon_type);
4370
4371 case BTF_KIND_FWD:
4372 case BTF_KIND_FLOAT:
4373 return btf_equal_common(cand_type, canon_type);
4374
4375 case BTF_KIND_CONST:
4376 case BTF_KIND_VOLATILE:
4377 case BTF_KIND_RESTRICT:
4378 case BTF_KIND_PTR:
4379 case BTF_KIND_TYPEDEF:
4380 case BTF_KIND_FUNC:
4381 case BTF_KIND_TYPE_TAG:
4382 if (cand_type->info != canon_type->info)
4383 return 0;
4384 return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4385
4386 case BTF_KIND_ARRAY: {
4387 const struct btf_array *cand_arr, *canon_arr;
4388
4389 if (!btf_compat_array(cand_type, canon_type))
4390 return 0;
4391 cand_arr = btf_array(cand_type);
4392 canon_arr = btf_array(canon_type);
4393 eq = btf_dedup_is_equiv(d, cand_arr->index_type, canon_arr->index_type);
4394 if (eq <= 0)
4395 return eq;
4396 return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
4397 }
4398
4399 case BTF_KIND_STRUCT:
4400 case BTF_KIND_UNION: {
4401 const struct btf_member *cand_m, *canon_m;
4402 __u16 vlen;
4403
4404 if (!btf_shallow_equal_struct(cand_type, canon_type))
4405 return 0;
4406 vlen = btf_vlen(cand_type);
4407 cand_m = btf_members(cand_type);
4408 canon_m = btf_members(canon_type);
4409 for (i = 0; i < vlen; i++) {
4410 eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
4411 if (eq <= 0)
4412 return eq;
4413 cand_m++;
4414 canon_m++;
4415 }
4416
4417 return 1;
4418 }
4419
4420 case BTF_KIND_FUNC_PROTO: {
4421 const struct btf_param *cand_p, *canon_p;
4422 __u16 vlen;
4423
4424 if (!btf_compat_fnproto(cand_type, canon_type))
4425 return 0;
4426 eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
4427 if (eq <= 0)
4428 return eq;
4429 vlen = btf_vlen(cand_type);
4430 cand_p = btf_params(cand_type);
4431 canon_p = btf_params(canon_type);
4432 for (i = 0; i < vlen; i++) {
4433 eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
4434 if (eq <= 0)
4435 return eq;
4436 cand_p++;
4437 canon_p++;
4438 }
4439 return 1;
4440 }
4441
4442 default:
4443 return -EINVAL;
4444 }
4445 return 0;
4446 }
4447
4448 /*
4449 * Use hypothetical mapping, produced by successful type graph equivalence
4450 * check, to augment existing struct/union canonical mapping, where possible.
4451 *
4452 * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
4453 * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
4454 * it doesn't matter if FWD type was part of canonical graph or candidate one,
4455 * we are recording the mapping anyway. As opposed to carefulness required
4456 * for struct/union correspondence mapping (described below), for FWD resolution
4457 * it's not important, as by the time that FWD type (reference type) will be
4458 * deduplicated all structs/unions will be deduped already anyway.
4459 *
4460 * Recording STRUCT/UNION mapping is purely a performance optimization and is
4461 * not required for correctness. It needs to be done carefully to ensure that
4462 * struct/union from candidate's type graph is not mapped into corresponding
4463 * struct/union from canonical type graph that itself hasn't been resolved into
4464 * canonical representative. The only guarantee we have is that canonical
4465 * struct/union was determined as canonical and that won't change. But any
4466 * types referenced through that struct/union fields could have been not yet
4467 * resolved, so in case like that it's too early to establish any kind of
4468 * correspondence between structs/unions.
4469 *
4470 * No canonical correspondence is derived for primitive types (they are already
4471 * deduplicated completely already anyway) or reference types (they rely on
4472 * stability of struct/union canonical relationship for equivalence checks).
4473 */
btf_dedup_merge_hypot_map(struct btf_dedup * d)4474 static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
4475 {
4476 __u32 canon_type_id, targ_type_id;
4477 __u16 t_kind, c_kind;
4478 __u32 t_id, c_id;
4479 int i;
4480
4481 for (i = 0; i < d->hypot_cnt; i++) {
4482 canon_type_id = d->hypot_list[i];
4483 targ_type_id = d->hypot_map[canon_type_id];
4484 t_id = resolve_type_id(d, targ_type_id);
4485 c_id = resolve_type_id(d, canon_type_id);
4486 t_kind = btf_kind(btf__type_by_id(d->btf, t_id));
4487 c_kind = btf_kind(btf__type_by_id(d->btf, c_id));
4488 /*
4489 * Resolve FWD into STRUCT/UNION.
4490 * It's ok to resolve FWD into STRUCT/UNION that's not yet
4491 * mapped to canonical representative (as opposed to
4492 * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
4493 * eventually that struct is going to be mapped and all resolved
4494 * FWDs will automatically resolve to correct canonical
4495 * representative. This will happen before ref type deduping,
4496 * which critically depends on stability of these mapping. This
4497 * stability is not a requirement for STRUCT/UNION equivalence
4498 * checks, though.
4499 */
4500
4501 /* if it's the split BTF case, we still need to point base FWD
4502 * to STRUCT/UNION in a split BTF, because FWDs from split BTF
4503 * will be resolved against base FWD. If we don't point base
4504 * canonical FWD to the resolved STRUCT/UNION, then all the
4505 * FWDs in split BTF won't be correctly resolved to a proper
4506 * STRUCT/UNION.
4507 */
4508 if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
4509 d->map[c_id] = t_id;
4510
4511 /* if graph equivalence determined that we'd need to adjust
4512 * base canonical types, then we need to only point base FWDs
4513 * to STRUCTs/UNIONs and do no more modifications. For all
4514 * other purposes the type graphs were not equivalent.
4515 */
4516 if (d->hypot_adjust_canon)
4517 continue;
4518
4519 if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
4520 d->map[t_id] = c_id;
4521
4522 if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
4523 c_kind != BTF_KIND_FWD &&
4524 is_type_mapped(d, c_id) &&
4525 !is_type_mapped(d, t_id)) {
4526 /*
4527 * as a perf optimization, we can map struct/union
4528 * that's part of type graph we just verified for
4529 * equivalence. We can do that for struct/union that has
4530 * canonical representative only, though.
4531 */
4532 d->map[t_id] = c_id;
4533 }
4534 }
4535 }
4536
4537 /*
4538 * Deduplicate struct/union types.
4539 *
4540 * For each struct/union type its type signature hash is calculated, taking
4541 * into account type's name, size, number, order and names of fields, but
4542 * ignoring type ID's referenced from fields, because they might not be deduped
4543 * completely until after reference types deduplication phase. This type hash
4544 * is used to iterate over all potential canonical types, sharing same hash.
4545 * For each canonical candidate we check whether type graphs that they form
4546 * (through referenced types in fields and so on) are equivalent using algorithm
4547 * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
4548 * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
4549 * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
4550 * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
4551 * potentially map other structs/unions to their canonical representatives,
4552 * if such relationship hasn't yet been established. This speeds up algorithm
4553 * by eliminating some of the duplicate work.
4554 *
4555 * If no matching canonical representative was found, struct/union is marked
4556 * as canonical for itself and is added into btf_dedup->dedup_table hash map
4557 * for further look ups.
4558 */
btf_dedup_struct_type(struct btf_dedup * d,__u32 type_id)4559 static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
4560 {
4561 struct btf_type *cand_type, *t;
4562 struct hashmap_entry *hash_entry;
4563 /* if we don't find equivalent type, then we are canonical */
4564 __u32 new_id = type_id;
4565 __u16 kind;
4566 long h;
4567
4568 /* already deduped or is in process of deduping (loop detected) */
4569 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4570 return 0;
4571
4572 t = btf_type_by_id(d->btf, type_id);
4573 kind = btf_kind(t);
4574
4575 if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4576 return 0;
4577
4578 h = btf_hash_struct(t);
4579 for_each_dedup_cand(d, hash_entry, h) {
4580 __u32 cand_id = hash_entry->value;
4581 int eq;
4582
4583 /*
4584 * Even though btf_dedup_is_equiv() checks for
4585 * btf_shallow_equal_struct() internally when checking two
4586 * structs (unions) for equivalence, we need to guard here
4587 * from picking matching FWD type as a dedup candidate.
4588 * This can happen due to hash collision. In such case just
4589 * relying on btf_dedup_is_equiv() would lead to potentially
4590 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
4591 * FWD and compatible STRUCT/UNION are considered equivalent.
4592 */
4593 cand_type = btf_type_by_id(d->btf, cand_id);
4594 if (!btf_shallow_equal_struct(t, cand_type))
4595 continue;
4596
4597 btf_dedup_clear_hypot_map(d);
4598 eq = btf_dedup_is_equiv(d, type_id, cand_id);
4599 if (eq < 0)
4600 return eq;
4601 if (!eq)
4602 continue;
4603 btf_dedup_merge_hypot_map(d);
4604 if (d->hypot_adjust_canon) /* not really equivalent */
4605 continue;
4606 new_id = cand_id;
4607 break;
4608 }
4609
4610 d->map[type_id] = new_id;
4611 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4612 return -ENOMEM;
4613
4614 return 0;
4615 }
4616
btf_dedup_struct_types(struct btf_dedup * d)4617 static int btf_dedup_struct_types(struct btf_dedup *d)
4618 {
4619 int i, err;
4620
4621 for (i = 0; i < d->btf->nr_types; i++) {
4622 err = btf_dedup_struct_type(d, d->btf->start_id + i);
4623 if (err)
4624 return err;
4625 }
4626 return 0;
4627 }
4628
4629 /*
4630 * Deduplicate reference type.
4631 *
4632 * Once all primitive and struct/union types got deduplicated, we can easily
4633 * deduplicate all other (reference) BTF types. This is done in two steps:
4634 *
4635 * 1. Resolve all referenced type IDs into their canonical type IDs. This
4636 * resolution can be done either immediately for primitive or struct/union types
4637 * (because they were deduped in previous two phases) or recursively for
4638 * reference types. Recursion will always terminate at either primitive or
4639 * struct/union type, at which point we can "unwind" chain of reference types
4640 * one by one. There is no danger of encountering cycles because in C type
4641 * system the only way to form type cycle is through struct/union, so any chain
4642 * of reference types, even those taking part in a type cycle, will inevitably
4643 * reach struct/union at some point.
4644 *
4645 * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
4646 * becomes "stable", in the sense that no further deduplication will cause
4647 * any changes to it. With that, it's now possible to calculate type's signature
4648 * hash (this time taking into account referenced type IDs) and loop over all
4649 * potential canonical representatives. If no match was found, current type
4650 * will become canonical representative of itself and will be added into
4651 * btf_dedup->dedup_table as another possible canonical representative.
4652 */
btf_dedup_ref_type(struct btf_dedup * d,__u32 type_id)4653 static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
4654 {
4655 struct hashmap_entry *hash_entry;
4656 __u32 new_id = type_id, cand_id;
4657 struct btf_type *t, *cand;
4658 /* if we don't find equivalent type, then we are representative type */
4659 int ref_type_id;
4660 long h;
4661
4662 if (d->map[type_id] == BTF_IN_PROGRESS_ID)
4663 return -ELOOP;
4664 if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4665 return resolve_type_id(d, type_id);
4666
4667 t = btf_type_by_id(d->btf, type_id);
4668 d->map[type_id] = BTF_IN_PROGRESS_ID;
4669
4670 switch (btf_kind(t)) {
4671 case BTF_KIND_CONST:
4672 case BTF_KIND_VOLATILE:
4673 case BTF_KIND_RESTRICT:
4674 case BTF_KIND_PTR:
4675 case BTF_KIND_TYPEDEF:
4676 case BTF_KIND_FUNC:
4677 case BTF_KIND_TYPE_TAG:
4678 ref_type_id = btf_dedup_ref_type(d, t->type);
4679 if (ref_type_id < 0)
4680 return ref_type_id;
4681 t->type = ref_type_id;
4682
4683 h = btf_hash_common(t);
4684 for_each_dedup_cand(d, hash_entry, h) {
4685 cand_id = hash_entry->value;
4686 cand = btf_type_by_id(d->btf, cand_id);
4687 if (btf_equal_common(t, cand)) {
4688 new_id = cand_id;
4689 break;
4690 }
4691 }
4692 break;
4693
4694 case BTF_KIND_DECL_TAG:
4695 ref_type_id = btf_dedup_ref_type(d, t->type);
4696 if (ref_type_id < 0)
4697 return ref_type_id;
4698 t->type = ref_type_id;
4699
4700 h = btf_hash_int_decl_tag(t);
4701 for_each_dedup_cand(d, hash_entry, h) {
4702 cand_id = hash_entry->value;
4703 cand = btf_type_by_id(d->btf, cand_id);
4704 if (btf_equal_int_tag(t, cand)) {
4705 new_id = cand_id;
4706 break;
4707 }
4708 }
4709 break;
4710
4711 case BTF_KIND_ARRAY: {
4712 struct btf_array *info = btf_array(t);
4713
4714 ref_type_id = btf_dedup_ref_type(d, info->type);
4715 if (ref_type_id < 0)
4716 return ref_type_id;
4717 info->type = ref_type_id;
4718
4719 ref_type_id = btf_dedup_ref_type(d, info->index_type);
4720 if (ref_type_id < 0)
4721 return ref_type_id;
4722 info->index_type = ref_type_id;
4723
4724 h = btf_hash_array(t);
4725 for_each_dedup_cand(d, hash_entry, h) {
4726 cand_id = hash_entry->value;
4727 cand = btf_type_by_id(d->btf, cand_id);
4728 if (btf_equal_array(t, cand)) {
4729 new_id = cand_id;
4730 break;
4731 }
4732 }
4733 break;
4734 }
4735
4736 case BTF_KIND_FUNC_PROTO: {
4737 struct btf_param *param;
4738 __u16 vlen;
4739 int i;
4740
4741 ref_type_id = btf_dedup_ref_type(d, t->type);
4742 if (ref_type_id < 0)
4743 return ref_type_id;
4744 t->type = ref_type_id;
4745
4746 vlen = btf_vlen(t);
4747 param = btf_params(t);
4748 for (i = 0; i < vlen; i++) {
4749 ref_type_id = btf_dedup_ref_type(d, param->type);
4750 if (ref_type_id < 0)
4751 return ref_type_id;
4752 param->type = ref_type_id;
4753 param++;
4754 }
4755
4756 h = btf_hash_fnproto(t);
4757 for_each_dedup_cand(d, hash_entry, h) {
4758 cand_id = hash_entry->value;
4759 cand = btf_type_by_id(d->btf, cand_id);
4760 if (btf_equal_fnproto(t, cand)) {
4761 new_id = cand_id;
4762 break;
4763 }
4764 }
4765 break;
4766 }
4767
4768 default:
4769 return -EINVAL;
4770 }
4771
4772 d->map[type_id] = new_id;
4773 if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4774 return -ENOMEM;
4775
4776 return new_id;
4777 }
4778
btf_dedup_ref_types(struct btf_dedup * d)4779 static int btf_dedup_ref_types(struct btf_dedup *d)
4780 {
4781 int i, err;
4782
4783 for (i = 0; i < d->btf->nr_types; i++) {
4784 err = btf_dedup_ref_type(d, d->btf->start_id + i);
4785 if (err < 0)
4786 return err;
4787 }
4788 /* we won't need d->dedup_table anymore */
4789 hashmap__free(d->dedup_table);
4790 d->dedup_table = NULL;
4791 return 0;
4792 }
4793
4794 /*
4795 * Collect a map from type names to type ids for all canonical structs
4796 * and unions. If the same name is shared by several canonical types
4797 * use a special value 0 to indicate this fact.
4798 */
btf_dedup_fill_unique_names_map(struct btf_dedup * d,struct hashmap * names_map)4799 static int btf_dedup_fill_unique_names_map(struct btf_dedup *d, struct hashmap *names_map)
4800 {
4801 __u32 nr_types = btf__type_cnt(d->btf);
4802 struct btf_type *t;
4803 __u32 type_id;
4804 __u16 kind;
4805 int err;
4806
4807 /*
4808 * Iterate over base and split module ids in order to get all
4809 * available structs in the map.
4810 */
4811 for (type_id = 1; type_id < nr_types; ++type_id) {
4812 t = btf_type_by_id(d->btf, type_id);
4813 kind = btf_kind(t);
4814
4815 if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4816 continue;
4817
4818 /* Skip non-canonical types */
4819 if (type_id != d->map[type_id])
4820 continue;
4821
4822 err = hashmap__add(names_map, t->name_off, type_id);
4823 if (err == -EEXIST)
4824 err = hashmap__set(names_map, t->name_off, 0, NULL, NULL);
4825
4826 if (err)
4827 return err;
4828 }
4829
4830 return 0;
4831 }
4832
btf_dedup_resolve_fwd(struct btf_dedup * d,struct hashmap * names_map,__u32 type_id)4833 static int btf_dedup_resolve_fwd(struct btf_dedup *d, struct hashmap *names_map, __u32 type_id)
4834 {
4835 struct btf_type *t = btf_type_by_id(d->btf, type_id);
4836 enum btf_fwd_kind fwd_kind = btf_kflag(t);
4837 __u16 cand_kind, kind = btf_kind(t);
4838 struct btf_type *cand_t;
4839 uintptr_t cand_id;
4840
4841 if (kind != BTF_KIND_FWD)
4842 return 0;
4843
4844 /* Skip if this FWD already has a mapping */
4845 if (type_id != d->map[type_id])
4846 return 0;
4847
4848 if (!hashmap__find(names_map, t->name_off, &cand_id))
4849 return 0;
4850
4851 /* Zero is a special value indicating that name is not unique */
4852 if (!cand_id)
4853 return 0;
4854
4855 cand_t = btf_type_by_id(d->btf, cand_id);
4856 cand_kind = btf_kind(cand_t);
4857 if ((cand_kind == BTF_KIND_STRUCT && fwd_kind != BTF_FWD_STRUCT) ||
4858 (cand_kind == BTF_KIND_UNION && fwd_kind != BTF_FWD_UNION))
4859 return 0;
4860
4861 d->map[type_id] = cand_id;
4862
4863 return 0;
4864 }
4865
4866 /*
4867 * Resolve unambiguous forward declarations.
4868 *
4869 * The lion's share of all FWD declarations is resolved during
4870 * `btf_dedup_struct_types` phase when different type graphs are
4871 * compared against each other. However, if in some compilation unit a
4872 * FWD declaration is not a part of a type graph compared against
4873 * another type graph that declaration's canonical type would not be
4874 * changed. Example:
4875 *
4876 * CU #1:
4877 *
4878 * struct foo;
4879 * struct foo *some_global;
4880 *
4881 * CU #2:
4882 *
4883 * struct foo { int u; };
4884 * struct foo *another_global;
4885 *
4886 * After `btf_dedup_struct_types` the BTF looks as follows:
4887 *
4888 * [1] STRUCT 'foo' size=4 vlen=1 ...
4889 * [2] INT 'int' size=4 ...
4890 * [3] PTR '(anon)' type_id=1
4891 * [4] FWD 'foo' fwd_kind=struct
4892 * [5] PTR '(anon)' type_id=4
4893 *
4894 * This pass assumes that such FWD declarations should be mapped to
4895 * structs or unions with identical name in case if the name is not
4896 * ambiguous.
4897 */
btf_dedup_resolve_fwds(struct btf_dedup * d)4898 static int btf_dedup_resolve_fwds(struct btf_dedup *d)
4899 {
4900 int i, err;
4901 struct hashmap *names_map;
4902
4903 names_map = hashmap__new(btf_dedup_identity_hash_fn, btf_dedup_equal_fn, NULL);
4904 if (IS_ERR(names_map))
4905 return PTR_ERR(names_map);
4906
4907 err = btf_dedup_fill_unique_names_map(d, names_map);
4908 if (err < 0)
4909 goto exit;
4910
4911 for (i = 0; i < d->btf->nr_types; i++) {
4912 err = btf_dedup_resolve_fwd(d, names_map, d->btf->start_id + i);
4913 if (err < 0)
4914 break;
4915 }
4916
4917 exit:
4918 hashmap__free(names_map);
4919 return err;
4920 }
4921
4922 /*
4923 * Compact types.
4924 *
4925 * After we established for each type its corresponding canonical representative
4926 * type, we now can eliminate types that are not canonical and leave only
4927 * canonical ones layed out sequentially in memory by copying them over
4928 * duplicates. During compaction btf_dedup->hypot_map array is reused to store
4929 * a map from original type ID to a new compacted type ID, which will be used
4930 * during next phase to "fix up" type IDs, referenced from struct/union and
4931 * reference types.
4932 */
btf_dedup_compact_types(struct btf_dedup * d)4933 static int btf_dedup_compact_types(struct btf_dedup *d)
4934 {
4935 __u32 *new_offs;
4936 __u32 next_type_id = d->btf->start_id;
4937 const struct btf_type *t;
4938 void *p;
4939 int i, id, len;
4940
4941 /* we are going to reuse hypot_map to store compaction remapping */
4942 d->hypot_map[0] = 0;
4943 /* base BTF types are not renumbered */
4944 for (id = 1; id < d->btf->start_id; id++)
4945 d->hypot_map[id] = id;
4946 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++)
4947 d->hypot_map[id] = BTF_UNPROCESSED_ID;
4948
4949 p = d->btf->types_data;
4950
4951 for (i = 0, id = d->btf->start_id; i < d->btf->nr_types; i++, id++) {
4952 if (d->map[id] != id)
4953 continue;
4954
4955 t = btf__type_by_id(d->btf, id);
4956 len = btf_type_size(t);
4957 if (len < 0)
4958 return len;
4959
4960 memmove(p, t, len);
4961 d->hypot_map[id] = next_type_id;
4962 d->btf->type_offs[next_type_id - d->btf->start_id] = p - d->btf->types_data;
4963 p += len;
4964 next_type_id++;
4965 }
4966
4967 /* shrink struct btf's internal types index and update btf_header */
4968 d->btf->nr_types = next_type_id - d->btf->start_id;
4969 d->btf->type_offs_cap = d->btf->nr_types;
4970 d->btf->hdr->type_len = p - d->btf->types_data;
4971 new_offs = libbpf_reallocarray(d->btf->type_offs, d->btf->type_offs_cap,
4972 sizeof(*new_offs));
4973 if (d->btf->type_offs_cap && !new_offs)
4974 return -ENOMEM;
4975 d->btf->type_offs = new_offs;
4976 d->btf->hdr->str_off = d->btf->hdr->type_len;
4977 d->btf->raw_size = d->btf->hdr->hdr_len + d->btf->hdr->type_len + d->btf->hdr->str_len;
4978 return 0;
4979 }
4980
4981 /*
4982 * Figure out final (deduplicated and compacted) type ID for provided original
4983 * `type_id` by first resolving it into corresponding canonical type ID and
4984 * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
4985 * which is populated during compaction phase.
4986 */
btf_dedup_remap_type_id(__u32 * type_id,void * ctx)4987 static int btf_dedup_remap_type_id(__u32 *type_id, void *ctx)
4988 {
4989 struct btf_dedup *d = ctx;
4990 __u32 resolved_type_id, new_type_id;
4991
4992 resolved_type_id = resolve_type_id(d, *type_id);
4993 new_type_id = d->hypot_map[resolved_type_id];
4994 if (new_type_id > BTF_MAX_NR_TYPES)
4995 return -EINVAL;
4996
4997 *type_id = new_type_id;
4998 return 0;
4999 }
5000
5001 /*
5002 * Remap referenced type IDs into deduped type IDs.
5003 *
5004 * After BTF types are deduplicated and compacted, their final type IDs may
5005 * differ from original ones. The map from original to a corresponding
5006 * deduped type ID is stored in btf_dedup->hypot_map and is populated during
5007 * compaction phase. During remapping phase we are rewriting all type IDs
5008 * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
5009 * their final deduped type IDs.
5010 */
btf_dedup_remap_types(struct btf_dedup * d)5011 static int btf_dedup_remap_types(struct btf_dedup *d)
5012 {
5013 int i, r;
5014
5015 for (i = 0; i < d->btf->nr_types; i++) {
5016 struct btf_type *t = btf_type_by_id(d->btf, d->btf->start_id + i);
5017 struct btf_field_iter it;
5018 __u32 *type_id;
5019
5020 r = btf_field_iter_init(&it, t, BTF_FIELD_ITER_IDS);
5021 if (r)
5022 return r;
5023
5024 while ((type_id = btf_field_iter_next(&it))) {
5025 __u32 resolved_id, new_id;
5026
5027 resolved_id = resolve_type_id(d, *type_id);
5028 new_id = d->hypot_map[resolved_id];
5029 if (new_id > BTF_MAX_NR_TYPES)
5030 return -EINVAL;
5031
5032 *type_id = new_id;
5033 }
5034 }
5035
5036 if (!d->btf_ext)
5037 return 0;
5038
5039 r = btf_ext_visit_type_ids(d->btf_ext, btf_dedup_remap_type_id, d);
5040 if (r)
5041 return r;
5042
5043 return 0;
5044 }
5045
5046 /*
5047 * Probe few well-known locations for vmlinux kernel image and try to load BTF
5048 * data out of it to use for target BTF.
5049 */
btf__load_vmlinux_btf(void)5050 struct btf *btf__load_vmlinux_btf(void)
5051 {
5052 const char *sysfs_btf_path = "/sys/kernel/btf/vmlinux";
5053 /* fall back locations, trying to find vmlinux on disk */
5054 const char *locations[] = {
5055 "/boot/vmlinux-%1$s",
5056 "/lib/modules/%1$s/vmlinux-%1$s",
5057 "/lib/modules/%1$s/build/vmlinux",
5058 "/usr/lib/modules/%1$s/kernel/vmlinux",
5059 "/usr/lib/debug/boot/vmlinux-%1$s",
5060 "/usr/lib/debug/boot/vmlinux-%1$s.debug",
5061 "/usr/lib/debug/lib/modules/%1$s/vmlinux",
5062 };
5063 char path[PATH_MAX + 1];
5064 struct utsname buf;
5065 struct btf *btf;
5066 int i, err;
5067
5068 /* is canonical sysfs location accessible? */
5069 if (faccessat(AT_FDCWD, sysfs_btf_path, F_OK, AT_EACCESS) < 0) {
5070 pr_warn("kernel BTF is missing at '%s', was CONFIG_DEBUG_INFO_BTF enabled?\n",
5071 sysfs_btf_path);
5072 } else {
5073 btf = btf__parse(sysfs_btf_path, NULL);
5074 if (!btf) {
5075 err = -errno;
5076 pr_warn("failed to read kernel BTF from '%s': %d\n", sysfs_btf_path, err);
5077 return libbpf_err_ptr(err);
5078 }
5079 pr_debug("loaded kernel BTF from '%s'\n", sysfs_btf_path);
5080 return btf;
5081 }
5082
5083 /* try fallback locations */
5084 uname(&buf);
5085 for (i = 0; i < ARRAY_SIZE(locations); i++) {
5086 snprintf(path, PATH_MAX, locations[i], buf.release);
5087
5088 if (faccessat(AT_FDCWD, path, R_OK, AT_EACCESS))
5089 continue;
5090
5091 btf = btf__parse(path, NULL);
5092 err = libbpf_get_error(btf);
5093 pr_debug("loading kernel BTF '%s': %d\n", path, err);
5094 if (err)
5095 continue;
5096
5097 return btf;
5098 }
5099
5100 pr_warn("failed to find valid kernel BTF\n");
5101 return libbpf_err_ptr(-ESRCH);
5102 }
5103
5104 struct btf *libbpf_find_kernel_btf(void) __attribute__((alias("btf__load_vmlinux_btf")));
5105
btf__load_module_btf(const char * module_name,struct btf * vmlinux_btf)5106 struct btf *btf__load_module_btf(const char *module_name, struct btf *vmlinux_btf)
5107 {
5108 char path[80];
5109
5110 snprintf(path, sizeof(path), "/sys/kernel/btf/%s", module_name);
5111 return btf__parse_split(path, vmlinux_btf);
5112 }
5113
btf_ext_visit_type_ids(struct btf_ext * btf_ext,type_id_visit_fn visit,void * ctx)5114 int btf_ext_visit_type_ids(struct btf_ext *btf_ext, type_id_visit_fn visit, void *ctx)
5115 {
5116 const struct btf_ext_info *seg;
5117 struct btf_ext_info_sec *sec;
5118 int i, err;
5119
5120 seg = &btf_ext->func_info;
5121 for_each_btf_ext_sec(seg, sec) {
5122 struct bpf_func_info_min *rec;
5123
5124 for_each_btf_ext_rec(seg, sec, i, rec) {
5125 err = visit(&rec->type_id, ctx);
5126 if (err < 0)
5127 return err;
5128 }
5129 }
5130
5131 seg = &btf_ext->core_relo_info;
5132 for_each_btf_ext_sec(seg, sec) {
5133 struct bpf_core_relo *rec;
5134
5135 for_each_btf_ext_rec(seg, sec, i, rec) {
5136 err = visit(&rec->type_id, ctx);
5137 if (err < 0)
5138 return err;
5139 }
5140 }
5141
5142 return 0;
5143 }
5144
btf_ext_visit_str_offs(struct btf_ext * btf_ext,str_off_visit_fn visit,void * ctx)5145 int btf_ext_visit_str_offs(struct btf_ext *btf_ext, str_off_visit_fn visit, void *ctx)
5146 {
5147 const struct btf_ext_info *seg;
5148 struct btf_ext_info_sec *sec;
5149 int i, err;
5150
5151 seg = &btf_ext->func_info;
5152 for_each_btf_ext_sec(seg, sec) {
5153 err = visit(&sec->sec_name_off, ctx);
5154 if (err)
5155 return err;
5156 }
5157
5158 seg = &btf_ext->line_info;
5159 for_each_btf_ext_sec(seg, sec) {
5160 struct bpf_line_info_min *rec;
5161
5162 err = visit(&sec->sec_name_off, ctx);
5163 if (err)
5164 return err;
5165
5166 for_each_btf_ext_rec(seg, sec, i, rec) {
5167 err = visit(&rec->file_name_off, ctx);
5168 if (err)
5169 return err;
5170 err = visit(&rec->line_off, ctx);
5171 if (err)
5172 return err;
5173 }
5174 }
5175
5176 seg = &btf_ext->core_relo_info;
5177 for_each_btf_ext_sec(seg, sec) {
5178 struct bpf_core_relo *rec;
5179
5180 err = visit(&sec->sec_name_off, ctx);
5181 if (err)
5182 return err;
5183
5184 for_each_btf_ext_rec(seg, sec, i, rec) {
5185 err = visit(&rec->access_str_off, ctx);
5186 if (err)
5187 return err;
5188 }
5189 }
5190
5191 return 0;
5192 }
5193
5194 struct btf_distill {
5195 struct btf_pipe pipe;
5196 int *id_map;
5197 unsigned int split_start_id;
5198 unsigned int split_start_str;
5199 int diff_id;
5200 };
5201
btf_add_distilled_type_ids(struct btf_distill * dist,__u32 i)5202 static int btf_add_distilled_type_ids(struct btf_distill *dist, __u32 i)
5203 {
5204 struct btf_type *split_t = btf_type_by_id(dist->pipe.src, i);
5205 struct btf_field_iter it;
5206 __u32 *id;
5207 int err;
5208
5209 err = btf_field_iter_init(&it, split_t, BTF_FIELD_ITER_IDS);
5210 if (err)
5211 return err;
5212 while ((id = btf_field_iter_next(&it))) {
5213 struct btf_type *base_t;
5214
5215 if (!*id)
5216 continue;
5217 /* split BTF id, not needed */
5218 if (*id >= dist->split_start_id)
5219 continue;
5220 /* already added ? */
5221 if (dist->id_map[*id] > 0)
5222 continue;
5223
5224 /* only a subset of base BTF types should be referenced from
5225 * split BTF; ensure nothing unexpected is referenced.
5226 */
5227 base_t = btf_type_by_id(dist->pipe.src, *id);
5228 switch (btf_kind(base_t)) {
5229 case BTF_KIND_INT:
5230 case BTF_KIND_FLOAT:
5231 case BTF_KIND_FWD:
5232 case BTF_KIND_ARRAY:
5233 case BTF_KIND_STRUCT:
5234 case BTF_KIND_UNION:
5235 case BTF_KIND_TYPEDEF:
5236 case BTF_KIND_ENUM:
5237 case BTF_KIND_ENUM64:
5238 case BTF_KIND_PTR:
5239 case BTF_KIND_CONST:
5240 case BTF_KIND_RESTRICT:
5241 case BTF_KIND_VOLATILE:
5242 case BTF_KIND_FUNC_PROTO:
5243 case BTF_KIND_TYPE_TAG:
5244 dist->id_map[*id] = *id;
5245 break;
5246 default:
5247 pr_warn("unexpected reference to base type[%u] of kind [%u] when creating distilled base BTF.\n",
5248 *id, btf_kind(base_t));
5249 return -EINVAL;
5250 }
5251 /* If a base type is used, ensure types it refers to are
5252 * marked as used also; so for example if we find a PTR to INT
5253 * we need both the PTR and INT.
5254 *
5255 * The only exception is named struct/unions, since distilled
5256 * base BTF composite types have no members.
5257 */
5258 if (btf_is_composite(base_t) && base_t->name_off)
5259 continue;
5260 err = btf_add_distilled_type_ids(dist, *id);
5261 if (err)
5262 return err;
5263 }
5264 return 0;
5265 }
5266
btf_add_distilled_types(struct btf_distill * dist)5267 static int btf_add_distilled_types(struct btf_distill *dist)
5268 {
5269 bool adding_to_base = dist->pipe.dst->start_id == 1;
5270 int id = btf__type_cnt(dist->pipe.dst);
5271 struct btf_type *t;
5272 int i, err = 0;
5273
5274
5275 /* Add types for each of the required references to either distilled
5276 * base or split BTF, depending on type characteristics.
5277 */
5278 for (i = 1; i < dist->split_start_id; i++) {
5279 const char *name;
5280 int kind;
5281
5282 if (!dist->id_map[i])
5283 continue;
5284 t = btf_type_by_id(dist->pipe.src, i);
5285 kind = btf_kind(t);
5286 name = btf__name_by_offset(dist->pipe.src, t->name_off);
5287
5288 switch (kind) {
5289 case BTF_KIND_INT:
5290 case BTF_KIND_FLOAT:
5291 case BTF_KIND_FWD:
5292 /* Named int, float, fwd are added to base. */
5293 if (!adding_to_base)
5294 continue;
5295 err = btf_add_type(&dist->pipe, t);
5296 break;
5297 case BTF_KIND_STRUCT:
5298 case BTF_KIND_UNION:
5299 /* Named struct/union are added to base as 0-vlen
5300 * struct/union of same size. Anonymous struct/unions
5301 * are added to split BTF as-is.
5302 */
5303 if (adding_to_base) {
5304 if (!t->name_off)
5305 continue;
5306 err = btf_add_composite(dist->pipe.dst, kind, name, t->size);
5307 } else {
5308 if (t->name_off)
5309 continue;
5310 err = btf_add_type(&dist->pipe, t);
5311 }
5312 break;
5313 case BTF_KIND_ENUM:
5314 case BTF_KIND_ENUM64:
5315 /* Named enum[64]s are added to base as a sized
5316 * enum; relocation will match with appropriately-named
5317 * and sized enum or enum64.
5318 *
5319 * Anonymous enums are added to split BTF as-is.
5320 */
5321 if (adding_to_base) {
5322 if (!t->name_off)
5323 continue;
5324 err = btf__add_enum(dist->pipe.dst, name, t->size);
5325 } else {
5326 if (t->name_off)
5327 continue;
5328 err = btf_add_type(&dist->pipe, t);
5329 }
5330 break;
5331 case BTF_KIND_ARRAY:
5332 case BTF_KIND_TYPEDEF:
5333 case BTF_KIND_PTR:
5334 case BTF_KIND_CONST:
5335 case BTF_KIND_RESTRICT:
5336 case BTF_KIND_VOLATILE:
5337 case BTF_KIND_FUNC_PROTO:
5338 case BTF_KIND_TYPE_TAG:
5339 /* All other types are added to split BTF. */
5340 if (adding_to_base)
5341 continue;
5342 err = btf_add_type(&dist->pipe, t);
5343 break;
5344 default:
5345 pr_warn("unexpected kind when adding base type '%s'[%u] of kind [%u] to distilled base BTF.\n",
5346 name, i, kind);
5347 return -EINVAL;
5348
5349 }
5350 if (err < 0)
5351 break;
5352 dist->id_map[i] = id++;
5353 }
5354 return err;
5355 }
5356
5357 /* Split BTF ids without a mapping will be shifted downwards since distilled
5358 * base BTF is smaller than the original base BTF. For those that have a
5359 * mapping (either to base or updated split BTF), update the id based on
5360 * that mapping.
5361 */
btf_update_distilled_type_ids(struct btf_distill * dist,__u32 i)5362 static int btf_update_distilled_type_ids(struct btf_distill *dist, __u32 i)
5363 {
5364 struct btf_type *t = btf_type_by_id(dist->pipe.dst, i);
5365 struct btf_field_iter it;
5366 __u32 *id;
5367 int err;
5368
5369 err = btf_field_iter_init(&it, t, BTF_FIELD_ITER_IDS);
5370 if (err)
5371 return err;
5372 while ((id = btf_field_iter_next(&it))) {
5373 if (dist->id_map[*id])
5374 *id = dist->id_map[*id];
5375 else if (*id >= dist->split_start_id)
5376 *id -= dist->diff_id;
5377 }
5378 return 0;
5379 }
5380
5381 /* Create updated split BTF with distilled base BTF; distilled base BTF
5382 * consists of BTF information required to clarify the types that split
5383 * BTF refers to, omitting unneeded details. Specifically it will contain
5384 * base types and memberless definitions of named structs, unions and enumerated
5385 * types. Associated reference types like pointers, arrays and anonymous
5386 * structs, unions and enumerated types will be added to split BTF.
5387 * Size is recorded for named struct/unions to help guide matching to the
5388 * target base BTF during later relocation.
5389 *
5390 * The only case where structs, unions or enumerated types are fully represented
5391 * is when they are anonymous; in such cases, the anonymous type is added to
5392 * split BTF in full.
5393 *
5394 * We return newly-created split BTF where the split BTF refers to a newly-created
5395 * distilled base BTF. Both must be freed separately by the caller.
5396 */
btf__distill_base(const struct btf * src_btf,struct btf ** new_base_btf,struct btf ** new_split_btf)5397 int btf__distill_base(const struct btf *src_btf, struct btf **new_base_btf,
5398 struct btf **new_split_btf)
5399 {
5400 struct btf *new_base = NULL, *new_split = NULL;
5401 const struct btf *old_base;
5402 unsigned int n = btf__type_cnt(src_btf);
5403 struct btf_distill dist = {};
5404 struct btf_type *t;
5405 int i, err = 0;
5406
5407 /* src BTF must be split BTF. */
5408 old_base = btf__base_btf(src_btf);
5409 if (!new_base_btf || !new_split_btf || !old_base)
5410 return libbpf_err(-EINVAL);
5411
5412 new_base = btf__new_empty();
5413 if (!new_base)
5414 return libbpf_err(-ENOMEM);
5415
5416 btf__set_endianness(new_base, btf__endianness(src_btf));
5417
5418 dist.id_map = calloc(n, sizeof(*dist.id_map));
5419 if (!dist.id_map) {
5420 err = -ENOMEM;
5421 goto done;
5422 }
5423 dist.pipe.src = src_btf;
5424 dist.pipe.dst = new_base;
5425 dist.pipe.str_off_map = hashmap__new(btf_dedup_identity_hash_fn, btf_dedup_equal_fn, NULL);
5426 if (IS_ERR(dist.pipe.str_off_map)) {
5427 err = -ENOMEM;
5428 goto done;
5429 }
5430 dist.split_start_id = btf__type_cnt(old_base);
5431 dist.split_start_str = old_base->hdr->str_len;
5432
5433 /* Pass over src split BTF; generate the list of base BTF type ids it
5434 * references; these will constitute our distilled BTF set to be
5435 * distributed over base and split BTF as appropriate.
5436 */
5437 for (i = src_btf->start_id; i < n; i++) {
5438 err = btf_add_distilled_type_ids(&dist, i);
5439 if (err < 0)
5440 goto done;
5441 }
5442 /* Next add types for each of the required references to base BTF and split BTF
5443 * in turn.
5444 */
5445 err = btf_add_distilled_types(&dist);
5446 if (err < 0)
5447 goto done;
5448
5449 /* Create new split BTF with distilled base BTF as its base; the final
5450 * state is split BTF with distilled base BTF that represents enough
5451 * about its base references to allow it to be relocated with the base
5452 * BTF available.
5453 */
5454 new_split = btf__new_empty_split(new_base);
5455 if (!new_split) {
5456 err = -errno;
5457 goto done;
5458 }
5459 dist.pipe.dst = new_split;
5460 /* First add all split types */
5461 for (i = src_btf->start_id; i < n; i++) {
5462 t = btf_type_by_id(src_btf, i);
5463 err = btf_add_type(&dist.pipe, t);
5464 if (err < 0)
5465 goto done;
5466 }
5467 /* Now add distilled types to split BTF that are not added to base. */
5468 err = btf_add_distilled_types(&dist);
5469 if (err < 0)
5470 goto done;
5471
5472 /* All split BTF ids will be shifted downwards since there are less base
5473 * BTF ids in distilled base BTF.
5474 */
5475 dist.diff_id = dist.split_start_id - btf__type_cnt(new_base);
5476
5477 n = btf__type_cnt(new_split);
5478 /* Now update base/split BTF ids. */
5479 for (i = 1; i < n; i++) {
5480 err = btf_update_distilled_type_ids(&dist, i);
5481 if (err < 0)
5482 break;
5483 }
5484 done:
5485 free(dist.id_map);
5486 hashmap__free(dist.pipe.str_off_map);
5487 if (err) {
5488 btf__free(new_split);
5489 btf__free(new_base);
5490 return libbpf_err(err);
5491 }
5492 *new_base_btf = new_base;
5493 *new_split_btf = new_split;
5494
5495 return 0;
5496 }
5497
btf_header(const struct btf * btf)5498 const struct btf_header *btf_header(const struct btf *btf)
5499 {
5500 return btf->hdr;
5501 }
5502
btf_set_base_btf(struct btf * btf,const struct btf * base_btf)5503 void btf_set_base_btf(struct btf *btf, const struct btf *base_btf)
5504 {
5505 btf->base_btf = (struct btf *)base_btf;
5506 btf->start_id = btf__type_cnt(base_btf);
5507 btf->start_str_off = base_btf->hdr->str_len;
5508 }
5509
btf__relocate(struct btf * btf,const struct btf * base_btf)5510 int btf__relocate(struct btf *btf, const struct btf *base_btf)
5511 {
5512 int err = btf_relocate(btf, base_btf, NULL);
5513
5514 if (!err)
5515 btf->owns_base = false;
5516 return libbpf_err(err);
5517 }
5518