hc
2023-12-08 01573e231f18eb2d99162747186f59511f56b64d
kernel/tools/lib/bpf/btf.c
....@@ -1,222 +1,534 @@
1
-// SPDX-License-Identifier: LGPL-2.1
1
+// SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
22 /* Copyright (c) 2018 Facebook */
33
4
+#include <byteswap.h>
5
+#include <endian.h>
6
+#include <stdio.h>
47 #include <stdlib.h>
58 #include <string.h>
9
+#include <fcntl.h>
610 #include <unistd.h>
711 #include <errno.h>
12
+#include <sys/utsname.h>
13
+#include <sys/param.h>
14
+#include <sys/stat.h>
15
+#include <linux/kernel.h>
816 #include <linux/err.h>
917 #include <linux/btf.h>
18
+#include <gelf.h>
1019 #include "btf.h"
1120 #include "bpf.h"
21
+#include "libbpf.h"
22
+#include "libbpf_internal.h"
23
+#include "hashmap.h"
1224
13
-#define elog(fmt, ...) { if (err_log) err_log(fmt, ##__VA_ARGS__); }
14
-#define max(a, b) ((a) > (b) ? (a) : (b))
15
-#define min(a, b) ((a) < (b) ? (a) : (b))
16
-
17
-#define BTF_MAX_NR_TYPES 65535
18
-
19
-#define IS_MODIFIER(k) (((k) == BTF_KIND_TYPEDEF) || \
20
- ((k) == BTF_KIND_VOLATILE) || \
21
- ((k) == BTF_KIND_CONST) || \
22
- ((k) == BTF_KIND_RESTRICT))
25
+#define BTF_MAX_NR_TYPES 0x7fffffffU
26
+#define BTF_MAX_STR_OFFSET 0x7fffffffU
2327
2428 static struct btf_type btf_void;
2529
2630 struct btf {
27
- union {
28
- struct btf_header *hdr;
29
- void *data;
30
- };
31
- struct btf_type **types;
32
- const char *strings;
33
- void *nohdr_data;
31
+ /* raw BTF data in native endianness */
32
+ void *raw_data;
33
+ /* raw BTF data in non-native endianness */
34
+ void *raw_data_swapped;
35
+ __u32 raw_size;
36
+ /* whether target endianness differs from the native one */
37
+ bool swapped_endian;
38
+
39
+ /*
40
+ * When BTF is loaded from an ELF or raw memory it is stored
41
+ * in a contiguous memory block. The hdr, type_data, and, strs_data
42
+ * point inside that memory region to their respective parts of BTF
43
+ * representation:
44
+ *
45
+ * +--------------------------------+
46
+ * | Header | Types | Strings |
47
+ * +--------------------------------+
48
+ * ^ ^ ^
49
+ * | | |
50
+ * hdr | |
51
+ * types_data-+ |
52
+ * strs_data------------+
53
+ *
54
+ * If BTF data is later modified, e.g., due to types added or
55
+ * removed, BTF deduplication performed, etc, this contiguous
56
+ * representation is broken up into three independently allocated
57
+ * memory regions to be able to modify them independently.
58
+ * raw_data is nulled out at that point, but can be later allocated
59
+ * and cached again if user calls btf__get_raw_data(), at which point
60
+ * raw_data will contain a contiguous copy of header, types, and
61
+ * strings:
62
+ *
63
+ * +----------+ +---------+ +-----------+
64
+ * | Header | | Types | | Strings |
65
+ * +----------+ +---------+ +-----------+
66
+ * ^ ^ ^
67
+ * | | |
68
+ * hdr | |
69
+ * types_data----+ |
70
+ * strs_data------------------+
71
+ *
72
+ * +----------+---------+-----------+
73
+ * | Header | Types | Strings |
74
+ * raw_data----->+----------+---------+-----------+
75
+ */
76
+ struct btf_header *hdr;
77
+
78
+ void *types_data;
79
+ size_t types_data_cap; /* used size stored in hdr->type_len */
80
+
81
+ /* type ID to `struct btf_type *` lookup index */
82
+ __u32 *type_offs;
83
+ size_t type_offs_cap;
3484 __u32 nr_types;
35
- __u32 types_size;
36
- __u32 data_size;
85
+
86
+ void *strs_data;
87
+ size_t strs_data_cap; /* used size stored in hdr->str_len */
88
+
89
+ /* lookup index for each unique string in strings section */
90
+ struct hashmap *strs_hash;
91
+ /* whether strings are already deduplicated */
92
+ bool strs_deduped;
93
+ /* BTF object FD, if loaded into kernel */
3794 int fd;
95
+
96
+ /* Pointer size (in bytes) for a target architecture of this BTF */
97
+ int ptr_sz;
3898 };
3999
40
-static int btf_add_type(struct btf *btf, struct btf_type *t)
100
+static inline __u64 ptr_to_u64(const void *ptr)
41101 {
42
- if (btf->types_size - btf->nr_types < 2) {
43
- struct btf_type **new_types;
44
- __u32 expand_by, new_size;
102
+ return (__u64) (unsigned long) ptr;
103
+}
45104
46
- if (btf->types_size == BTF_MAX_NR_TYPES)
47
- return -E2BIG;
105
+/* Ensure given dynamically allocated memory region pointed to by *data* with
106
+ * capacity of *cap_cnt* elements each taking *elem_sz* bytes has enough
107
+ * memory to accomodate *add_cnt* new elements, assuming *cur_cnt* elements
108
+ * are already used. At most *max_cnt* elements can be ever allocated.
109
+ * If necessary, memory is reallocated and all existing data is copied over,
110
+ * new pointer to the memory region is stored at *data, new memory region
111
+ * capacity (in number of elements) is stored in *cap.
112
+ * On success, memory pointer to the beginning of unused memory is returned.
113
+ * On error, NULL is returned.
114
+ */
115
+void *btf_add_mem(void **data, size_t *cap_cnt, size_t elem_sz,
116
+ size_t cur_cnt, size_t max_cnt, size_t add_cnt)
117
+{
118
+ size_t new_cnt;
119
+ void *new_data;
48120
49
- expand_by = max(btf->types_size >> 2, 16);
50
- new_size = min(BTF_MAX_NR_TYPES, btf->types_size + expand_by);
121
+ if (cur_cnt + add_cnt <= *cap_cnt)
122
+ return *data + cur_cnt * elem_sz;
51123
52
- new_types = realloc(btf->types, sizeof(*new_types) * new_size);
53
- if (!new_types)
54
- return -ENOMEM;
124
+ /* requested more than the set limit */
125
+ if (cur_cnt + add_cnt > max_cnt)
126
+ return NULL;
55127
56
- if (btf->nr_types == 0)
57
- new_types[0] = &btf_void;
128
+ new_cnt = *cap_cnt;
129
+ new_cnt += new_cnt / 4; /* expand by 25% */
130
+ if (new_cnt < 16) /* but at least 16 elements */
131
+ new_cnt = 16;
132
+ if (new_cnt > max_cnt) /* but not exceeding a set limit */
133
+ new_cnt = max_cnt;
134
+ if (new_cnt < cur_cnt + add_cnt) /* also ensure we have enough memory */
135
+ new_cnt = cur_cnt + add_cnt;
58136
59
- btf->types = new_types;
60
- btf->types_size = new_size;
61
- }
137
+ new_data = libbpf_reallocarray(*data, new_cnt, elem_sz);
138
+ if (!new_data)
139
+ return NULL;
62140
63
- btf->types[++(btf->nr_types)] = t;
141
+ /* zero out newly allocated portion of memory */
142
+ memset(new_data + (*cap_cnt) * elem_sz, 0, (new_cnt - *cap_cnt) * elem_sz);
143
+
144
+ *data = new_data;
145
+ *cap_cnt = new_cnt;
146
+ return new_data + cur_cnt * elem_sz;
147
+}
148
+
149
+/* Ensure given dynamically allocated memory region has enough allocated space
150
+ * to accommodate *need_cnt* elements of size *elem_sz* bytes each
151
+ */
152
+int btf_ensure_mem(void **data, size_t *cap_cnt, size_t elem_sz, size_t need_cnt)
153
+{
154
+ void *p;
155
+
156
+ if (need_cnt <= *cap_cnt)
157
+ return 0;
158
+
159
+ p = btf_add_mem(data, cap_cnt, elem_sz, *cap_cnt, SIZE_MAX, need_cnt - *cap_cnt);
160
+ if (!p)
161
+ return -ENOMEM;
64162
65163 return 0;
66164 }
67165
68
-static int btf_parse_hdr(struct btf *btf, btf_print_fn_t err_log)
166
+static int btf_add_type_idx_entry(struct btf *btf, __u32 type_off)
69167 {
70
- const struct btf_header *hdr = btf->hdr;
71
- __u32 meta_left;
168
+ __u32 *p;
72169
73
- if (btf->data_size < sizeof(struct btf_header)) {
74
- elog("BTF header not found\n");
75
- return -EINVAL;
76
- }
170
+ p = btf_add_mem((void **)&btf->type_offs, &btf->type_offs_cap, sizeof(__u32),
171
+ btf->nr_types + 1, BTF_MAX_NR_TYPES, 1);
172
+ if (!p)
173
+ return -ENOMEM;
77174
78
- if (hdr->magic != BTF_MAGIC) {
79
- elog("Invalid BTF magic:%x\n", hdr->magic);
80
- return -EINVAL;
81
- }
82
-
83
- if (hdr->version != BTF_VERSION) {
84
- elog("Unsupported BTF version:%u\n", hdr->version);
85
- return -ENOTSUP;
86
- }
87
-
88
- if (hdr->flags) {
89
- elog("Unsupported BTF flags:%x\n", hdr->flags);
90
- return -ENOTSUP;
91
- }
92
-
93
- meta_left = btf->data_size - sizeof(*hdr);
94
- if (!meta_left) {
95
- elog("BTF has no data\n");
96
- return -EINVAL;
97
- }
98
-
99
- if (meta_left < hdr->type_off) {
100
- elog("Invalid BTF type section offset:%u\n", hdr->type_off);
101
- return -EINVAL;
102
- }
103
-
104
- if (meta_left < hdr->str_off) {
105
- elog("Invalid BTF string section offset:%u\n", hdr->str_off);
106
- return -EINVAL;
107
- }
108
-
109
- if (hdr->type_off >= hdr->str_off) {
110
- elog("BTF type section offset >= string section offset. No type?\n");
111
- return -EINVAL;
112
- }
113
-
114
- if (hdr->type_off & 0x02) {
115
- elog("BTF type section is not aligned to 4 bytes\n");
116
- return -EINVAL;
117
- }
118
-
119
- btf->nohdr_data = btf->hdr + 1;
120
-
175
+ *p = type_off;
121176 return 0;
122177 }
123178
124
-static int btf_parse_str_sec(struct btf *btf, btf_print_fn_t err_log)
179
+static void btf_bswap_hdr(struct btf_header *h)
125180 {
126
- const struct btf_header *hdr = btf->hdr;
127
- const char *start = btf->nohdr_data + hdr->str_off;
128
- const char *end = start + btf->hdr->str_len;
129
-
130
- if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_NAME_OFFSET ||
131
- start[0] || end[-1]) {
132
- elog("Invalid BTF string section\n");
133
- return -EINVAL;
134
- }
135
-
136
- btf->strings = start;
137
-
138
- return 0;
181
+ h->magic = bswap_16(h->magic);
182
+ h->hdr_len = bswap_32(h->hdr_len);
183
+ h->type_off = bswap_32(h->type_off);
184
+ h->type_len = bswap_32(h->type_len);
185
+ h->str_off = bswap_32(h->str_off);
186
+ h->str_len = bswap_32(h->str_len);
139187 }
140188
141
-static int btf_parse_type_sec(struct btf *btf, btf_print_fn_t err_log)
189
+static int btf_parse_hdr(struct btf *btf)
142190 {
143191 struct btf_header *hdr = btf->hdr;
144
- void *nohdr_data = btf->nohdr_data;
145
- void *next_type = nohdr_data + hdr->type_off;
146
- void *end_type = nohdr_data + hdr->str_off;
192
+ __u32 meta_left;
147193
148
- while (next_type < end_type) {
149
- struct btf_type *t = next_type;
150
- __u16 vlen = BTF_INFO_VLEN(t->info);
151
- int err;
194
+ if (btf->raw_size < sizeof(struct btf_header)) {
195
+ pr_debug("BTF header not found\n");
196
+ return -EINVAL;
197
+ }
152198
153
- next_type += sizeof(*t);
154
- switch (BTF_INFO_KIND(t->info)) {
155
- case BTF_KIND_INT:
156
- next_type += sizeof(int);
157
- break;
158
- case BTF_KIND_ARRAY:
159
- next_type += sizeof(struct btf_array);
160
- break;
161
- case BTF_KIND_STRUCT:
162
- case BTF_KIND_UNION:
163
- next_type += vlen * sizeof(struct btf_member);
164
- break;
165
- case BTF_KIND_ENUM:
166
- next_type += vlen * sizeof(struct btf_enum);
167
- break;
168
- case BTF_KIND_TYPEDEF:
169
- case BTF_KIND_PTR:
170
- case BTF_KIND_FWD:
171
- case BTF_KIND_VOLATILE:
172
- case BTF_KIND_CONST:
173
- case BTF_KIND_RESTRICT:
174
- break;
175
- default:
176
- elog("Unsupported BTF_KIND:%u\n",
177
- BTF_INFO_KIND(t->info));
199
+ if (hdr->magic == bswap_16(BTF_MAGIC)) {
200
+ btf->swapped_endian = true;
201
+ if (bswap_32(hdr->hdr_len) != sizeof(struct btf_header)) {
202
+ pr_warn("Can't load BTF with non-native endianness due to unsupported header length %u\n",
203
+ bswap_32(hdr->hdr_len));
204
+ return -ENOTSUP;
205
+ }
206
+ btf_bswap_hdr(hdr);
207
+ } else if (hdr->magic != BTF_MAGIC) {
208
+ pr_debug("Invalid BTF magic: %x\n", hdr->magic);
209
+ return -EINVAL;
210
+ }
211
+
212
+ if (btf->raw_size < hdr->hdr_len) {
213
+ pr_debug("BTF header len %u larger than data size %u\n",
214
+ hdr->hdr_len, btf->raw_size);
215
+ return -EINVAL;
216
+ }
217
+
218
+ meta_left = btf->raw_size - hdr->hdr_len;
219
+ if (meta_left < (long long)hdr->str_off + hdr->str_len) {
220
+ pr_debug("Invalid BTF total size: %u\n", btf->raw_size);
221
+ return -EINVAL;
222
+ }
223
+
224
+ if ((long long)hdr->type_off + hdr->type_len > hdr->str_off) {
225
+ pr_debug("Invalid BTF data sections layout: type data at %u + %u, strings data at %u + %u\n",
226
+ hdr->type_off, hdr->type_len, hdr->str_off, hdr->str_len);
227
+ return -EINVAL;
228
+ }
229
+
230
+ if (hdr->type_off % 4) {
231
+ pr_debug("BTF type section is not aligned to 4 bytes\n");
232
+ return -EINVAL;
233
+ }
234
+
235
+ return 0;
236
+}
237
+
238
+static int btf_parse_str_sec(struct btf *btf)
239
+{
240
+ const struct btf_header *hdr = btf->hdr;
241
+ const char *start = btf->strs_data;
242
+ const char *end = start + btf->hdr->str_len;
243
+
244
+ if (!hdr->str_len || hdr->str_len - 1 > BTF_MAX_STR_OFFSET ||
245
+ start[0] || end[-1]) {
246
+ pr_debug("Invalid BTF string section\n");
247
+ return -EINVAL;
248
+ }
249
+
250
+ return 0;
251
+}
252
+
253
+static int btf_type_size(const struct btf_type *t)
254
+{
255
+ const int base_size = sizeof(struct btf_type);
256
+ __u16 vlen = btf_vlen(t);
257
+
258
+ switch (btf_kind(t)) {
259
+ case BTF_KIND_FWD:
260
+ case BTF_KIND_CONST:
261
+ case BTF_KIND_VOLATILE:
262
+ case BTF_KIND_RESTRICT:
263
+ case BTF_KIND_PTR:
264
+ case BTF_KIND_TYPEDEF:
265
+ case BTF_KIND_FUNC:
266
+ return base_size;
267
+ case BTF_KIND_INT:
268
+ return base_size + sizeof(__u32);
269
+ case BTF_KIND_ENUM:
270
+ return base_size + vlen * sizeof(struct btf_enum);
271
+ case BTF_KIND_ARRAY:
272
+ return base_size + sizeof(struct btf_array);
273
+ case BTF_KIND_STRUCT:
274
+ case BTF_KIND_UNION:
275
+ return base_size + vlen * sizeof(struct btf_member);
276
+ case BTF_KIND_FUNC_PROTO:
277
+ return base_size + vlen * sizeof(struct btf_param);
278
+ case BTF_KIND_VAR:
279
+ return base_size + sizeof(struct btf_var);
280
+ case BTF_KIND_DATASEC:
281
+ return base_size + vlen * sizeof(struct btf_var_secinfo);
282
+ default:
283
+ pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
284
+ return -EINVAL;
285
+ }
286
+}
287
+
288
+static void btf_bswap_type_base(struct btf_type *t)
289
+{
290
+ t->name_off = bswap_32(t->name_off);
291
+ t->info = bswap_32(t->info);
292
+ t->type = bswap_32(t->type);
293
+}
294
+
295
+static int btf_bswap_type_rest(struct btf_type *t)
296
+{
297
+ struct btf_var_secinfo *v;
298
+ struct btf_member *m;
299
+ struct btf_array *a;
300
+ struct btf_param *p;
301
+ struct btf_enum *e;
302
+ __u16 vlen = btf_vlen(t);
303
+ int i;
304
+
305
+ switch (btf_kind(t)) {
306
+ case BTF_KIND_FWD:
307
+ case BTF_KIND_CONST:
308
+ case BTF_KIND_VOLATILE:
309
+ case BTF_KIND_RESTRICT:
310
+ case BTF_KIND_PTR:
311
+ case BTF_KIND_TYPEDEF:
312
+ case BTF_KIND_FUNC:
313
+ return 0;
314
+ case BTF_KIND_INT:
315
+ *(__u32 *)(t + 1) = bswap_32(*(__u32 *)(t + 1));
316
+ return 0;
317
+ case BTF_KIND_ENUM:
318
+ for (i = 0, e = btf_enum(t); i < vlen; i++, e++) {
319
+ e->name_off = bswap_32(e->name_off);
320
+ e->val = bswap_32(e->val);
321
+ }
322
+ return 0;
323
+ case BTF_KIND_ARRAY:
324
+ a = btf_array(t);
325
+ a->type = bswap_32(a->type);
326
+ a->index_type = bswap_32(a->index_type);
327
+ a->nelems = bswap_32(a->nelems);
328
+ return 0;
329
+ case BTF_KIND_STRUCT:
330
+ case BTF_KIND_UNION:
331
+ for (i = 0, m = btf_members(t); i < vlen; i++, m++) {
332
+ m->name_off = bswap_32(m->name_off);
333
+ m->type = bswap_32(m->type);
334
+ m->offset = bswap_32(m->offset);
335
+ }
336
+ return 0;
337
+ case BTF_KIND_FUNC_PROTO:
338
+ for (i = 0, p = btf_params(t); i < vlen; i++, p++) {
339
+ p->name_off = bswap_32(p->name_off);
340
+ p->type = bswap_32(p->type);
341
+ }
342
+ return 0;
343
+ case BTF_KIND_VAR:
344
+ btf_var(t)->linkage = bswap_32(btf_var(t)->linkage);
345
+ return 0;
346
+ case BTF_KIND_DATASEC:
347
+ for (i = 0, v = btf_var_secinfos(t); i < vlen; i++, v++) {
348
+ v->type = bswap_32(v->type);
349
+ v->offset = bswap_32(v->offset);
350
+ v->size = bswap_32(v->size);
351
+ }
352
+ return 0;
353
+ default:
354
+ pr_debug("Unsupported BTF_KIND:%u\n", btf_kind(t));
355
+ return -EINVAL;
356
+ }
357
+}
358
+
359
+static int btf_parse_type_sec(struct btf *btf)
360
+{
361
+ struct btf_header *hdr = btf->hdr;
362
+ void *next_type = btf->types_data;
363
+ void *end_type = next_type + hdr->type_len;
364
+ int err, i = 0, type_size;
365
+
366
+ /* VOID (type_id == 0) is specially handled by btf__get_type_by_id(),
367
+ * so ensure we can never properly use its offset from index by
368
+ * setting it to a large value
369
+ */
370
+ err = btf_add_type_idx_entry(btf, UINT_MAX);
371
+ if (err)
372
+ return err;
373
+
374
+ while (next_type + sizeof(struct btf_type) <= end_type) {
375
+ i++;
376
+
377
+ if (btf->swapped_endian)
378
+ btf_bswap_type_base(next_type);
379
+
380
+ type_size = btf_type_size(next_type);
381
+ if (type_size < 0)
382
+ return type_size;
383
+ if (next_type + type_size > end_type) {
384
+ pr_warn("BTF type [%d] is malformed\n", i);
178385 return -EINVAL;
179386 }
180387
181
- err = btf_add_type(btf, t);
388
+ if (btf->swapped_endian && btf_bswap_type_rest(next_type))
389
+ return -EINVAL;
390
+
391
+ err = btf_add_type_idx_entry(btf, next_type - btf->types_data);
182392 if (err)
183393 return err;
394
+
395
+ next_type += type_size;
396
+ btf->nr_types++;
397
+ }
398
+
399
+ if (next_type != end_type) {
400
+ pr_warn("BTF types data is malformed\n");
401
+ return -EINVAL;
184402 }
185403
186404 return 0;
405
+}
406
+
407
+__u32 btf__get_nr_types(const struct btf *btf)
408
+{
409
+ return btf->nr_types;
410
+}
411
+
412
+/* internal helper returning non-const pointer to a type */
413
+static struct btf_type *btf_type_by_id(struct btf *btf, __u32 type_id)
414
+{
415
+ if (type_id == 0)
416
+ return &btf_void;
417
+
418
+ return btf->types_data + btf->type_offs[type_id];
187419 }
188420
189421 const struct btf_type *btf__type_by_id(const struct btf *btf, __u32 type_id)
190422 {
191423 if (type_id > btf->nr_types)
192424 return NULL;
425
+ return btf_type_by_id((struct btf *)btf, type_id);
426
+}
193427
194
- return btf->types[type_id];
428
+static int determine_ptr_size(const struct btf *btf)
429
+{
430
+ const struct btf_type *t;
431
+ const char *name;
432
+ int i;
433
+
434
+ for (i = 1; i <= btf->nr_types; i++) {
435
+ t = btf__type_by_id(btf, i);
436
+ if (!btf_is_int(t))
437
+ continue;
438
+
439
+ name = btf__name_by_offset(btf, t->name_off);
440
+ if (!name)
441
+ continue;
442
+
443
+ if (strcmp(name, "long int") == 0 ||
444
+ strcmp(name, "long unsigned int") == 0) {
445
+ if (t->size != 4 && t->size != 8)
446
+ continue;
447
+ return t->size;
448
+ }
449
+ }
450
+
451
+ return -1;
452
+}
453
+
454
+static size_t btf_ptr_sz(const struct btf *btf)
455
+{
456
+ if (!btf->ptr_sz)
457
+ ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
458
+ return btf->ptr_sz < 0 ? sizeof(void *) : btf->ptr_sz;
459
+}
460
+
461
+/* Return pointer size this BTF instance assumes. The size is heuristically
462
+ * determined by looking for 'long' or 'unsigned long' integer type and
463
+ * recording its size in bytes. If BTF type information doesn't have any such
464
+ * type, this function returns 0. In the latter case, native architecture's
465
+ * pointer size is assumed, so will be either 4 or 8, depending on
466
+ * architecture that libbpf was compiled for. It's possible to override
467
+ * guessed value by using btf__set_pointer_size() API.
468
+ */
469
+size_t btf__pointer_size(const struct btf *btf)
470
+{
471
+ if (!btf->ptr_sz)
472
+ ((struct btf *)btf)->ptr_sz = determine_ptr_size(btf);
473
+
474
+ if (btf->ptr_sz < 0)
475
+ /* not enough BTF type info to guess */
476
+ return 0;
477
+
478
+ return btf->ptr_sz;
479
+}
480
+
481
+/* Override or set pointer size in bytes. Only values of 4 and 8 are
482
+ * supported.
483
+ */
484
+int btf__set_pointer_size(struct btf *btf, size_t ptr_sz)
485
+{
486
+ if (ptr_sz != 4 && ptr_sz != 8)
487
+ return -EINVAL;
488
+ btf->ptr_sz = ptr_sz;
489
+ return 0;
490
+}
491
+
492
+static bool is_host_big_endian(void)
493
+{
494
+#if __BYTE_ORDER == __LITTLE_ENDIAN
495
+ return false;
496
+#elif __BYTE_ORDER == __BIG_ENDIAN
497
+ return true;
498
+#else
499
+# error "Unrecognized __BYTE_ORDER__"
500
+#endif
501
+}
502
+
503
+enum btf_endianness btf__endianness(const struct btf *btf)
504
+{
505
+ if (is_host_big_endian())
506
+ return btf->swapped_endian ? BTF_LITTLE_ENDIAN : BTF_BIG_ENDIAN;
507
+ else
508
+ return btf->swapped_endian ? BTF_BIG_ENDIAN : BTF_LITTLE_ENDIAN;
509
+}
510
+
511
+int btf__set_endianness(struct btf *btf, enum btf_endianness endian)
512
+{
513
+ if (endian != BTF_LITTLE_ENDIAN && endian != BTF_BIG_ENDIAN)
514
+ return -EINVAL;
515
+
516
+ btf->swapped_endian = is_host_big_endian() != (endian == BTF_BIG_ENDIAN);
517
+ if (!btf->swapped_endian) {
518
+ free(btf->raw_data_swapped);
519
+ btf->raw_data_swapped = NULL;
520
+ }
521
+ return 0;
195522 }
196523
197524 static bool btf_type_is_void(const struct btf_type *t)
198525 {
199
- return t == &btf_void || BTF_INFO_KIND(t->info) == BTF_KIND_FWD;
526
+ return t == &btf_void || btf_is_fwd(t);
200527 }
201528
202529 static bool btf_type_is_void_or_null(const struct btf_type *t)
203530 {
204531 return !t || btf_type_is_void(t);
205
-}
206
-
207
-static __s64 btf_type_size(const struct btf_type *t)
208
-{
209
- switch (BTF_INFO_KIND(t->info)) {
210
- case BTF_KIND_INT:
211
- case BTF_KIND_STRUCT:
212
- case BTF_KIND_UNION:
213
- case BTF_KIND_ENUM:
214
- return t->size;
215
- case BTF_KIND_PTR:
216
- return sizeof(void *);
217
- default:
218
- return -EINVAL;
219
- }
220532 }
221533
222534 #define MAX_RESOLVE_DEPTH 32
....@@ -232,19 +544,26 @@
232544 t = btf__type_by_id(btf, type_id);
233545 for (i = 0; i < MAX_RESOLVE_DEPTH && !btf_type_is_void_or_null(t);
234546 i++) {
235
- size = btf_type_size(t);
236
- if (size >= 0)
237
- break;
238
-
239
- switch (BTF_INFO_KIND(t->info)) {
547
+ switch (btf_kind(t)) {
548
+ case BTF_KIND_INT:
549
+ case BTF_KIND_STRUCT:
550
+ case BTF_KIND_UNION:
551
+ case BTF_KIND_ENUM:
552
+ case BTF_KIND_DATASEC:
553
+ size = t->size;
554
+ goto done;
555
+ case BTF_KIND_PTR:
556
+ size = btf_ptr_sz(btf);
557
+ goto done;
240558 case BTF_KIND_TYPEDEF:
241559 case BTF_KIND_VOLATILE:
242560 case BTF_KIND_CONST:
243561 case BTF_KIND_RESTRICT:
562
+ case BTF_KIND_VAR:
244563 type_id = t->type;
245564 break;
246565 case BTF_KIND_ARRAY:
247
- array = (const struct btf_array *)(t + 1);
566
+ array = btf_array(t);
248567 if (nelems && array->nelems > UINT32_MAX / nelems)
249568 return -E2BIG;
250569 nelems *= array->nelems;
....@@ -257,13 +576,52 @@
257576 t = btf__type_by_id(btf, type_id);
258577 }
259578
579
+done:
260580 if (size < 0)
261581 return -EINVAL;
262
-
263582 if (nelems && size > UINT32_MAX / nelems)
264583 return -E2BIG;
265584
266585 return nelems * size;
586
+}
587
+
588
+int btf__align_of(const struct btf *btf, __u32 id)
589
+{
590
+ const struct btf_type *t = btf__type_by_id(btf, id);
591
+ __u16 kind = btf_kind(t);
592
+
593
+ switch (kind) {
594
+ case BTF_KIND_INT:
595
+ case BTF_KIND_ENUM:
596
+ return min(btf_ptr_sz(btf), (size_t)t->size);
597
+ case BTF_KIND_PTR:
598
+ return btf_ptr_sz(btf);
599
+ case BTF_KIND_TYPEDEF:
600
+ case BTF_KIND_VOLATILE:
601
+ case BTF_KIND_CONST:
602
+ case BTF_KIND_RESTRICT:
603
+ return btf__align_of(btf, t->type);
604
+ case BTF_KIND_ARRAY:
605
+ return btf__align_of(btf, btf_array(t)->type);
606
+ case BTF_KIND_STRUCT:
607
+ case BTF_KIND_UNION: {
608
+ const struct btf_member *m = btf_members(t);
609
+ __u16 vlen = btf_vlen(t);
610
+ int i, max_align = 1, align;
611
+
612
+ for (i = 0; i < vlen; i++, m++) {
613
+ align = btf__align_of(btf, m->type);
614
+ if (align <= 0)
615
+ return align;
616
+ max_align = max(max_align, align);
617
+ }
618
+
619
+ return max_align;
620
+ }
621
+ default:
622
+ pr_warn("unsupported BTF_KIND:%u\n", btf_kind(t));
623
+ return 0;
624
+ }
267625 }
268626
269627 int btf__resolve_type(const struct btf *btf, __u32 type_id)
....@@ -274,7 +632,7 @@
274632 t = btf__type_by_id(btf, type_id);
275633 while (depth < MAX_RESOLVE_DEPTH &&
276634 !btf_type_is_void_or_null(t) &&
277
- IS_MODIFIER(BTF_INFO_KIND(t->info))) {
635
+ (btf_is_mod(t) || btf_is_typedef(t) || btf_is_var(t))) {
278636 type_id = t->type;
279637 t = btf__type_by_id(btf, type_id);
280638 depth++;
....@@ -294,7 +652,7 @@
294652 return 0;
295653
296654 for (i = 1; i <= btf->nr_types; i++) {
297
- const struct btf_type *t = btf->types[i];
655
+ const struct btf_type *t = btf__type_by_id(btf, i);
298656 const char *name = btf__name_by_offset(btf, t->name_off);
299657
300658 if (name && !strcmp(type_name, name))
....@@ -304,23 +662,92 @@
304662 return -ENOENT;
305663 }
306664
665
+__s32 btf__find_by_name_kind(const struct btf *btf, const char *type_name,
666
+ __u32 kind)
667
+{
668
+ __u32 i;
669
+
670
+ if (kind == BTF_KIND_UNKN || !strcmp(type_name, "void"))
671
+ return 0;
672
+
673
+ for (i = 1; i <= btf->nr_types; i++) {
674
+ const struct btf_type *t = btf__type_by_id(btf, i);
675
+ const char *name;
676
+
677
+ if (btf_kind(t) != kind)
678
+ continue;
679
+ name = btf__name_by_offset(btf, t->name_off);
680
+ if (name && !strcmp(type_name, name))
681
+ return i;
682
+ }
683
+
684
+ return -ENOENT;
685
+}
686
+
687
+static bool btf_is_modifiable(const struct btf *btf)
688
+{
689
+ return (void *)btf->hdr != btf->raw_data;
690
+}
691
+
307692 void btf__free(struct btf *btf)
308693 {
309
- if (!btf)
694
+ if (IS_ERR_OR_NULL(btf))
310695 return;
311696
312
- if (btf->fd != -1)
697
+ if (btf->fd >= 0)
313698 close(btf->fd);
314699
315
- free(btf->data);
316
- free(btf->types);
700
+ if (btf_is_modifiable(btf)) {
701
+ /* if BTF was modified after loading, it will have a split
702
+ * in-memory representation for header, types, and strings
703
+ * sections, so we need to free all of them individually. It
704
+ * might still have a cached contiguous raw data present,
705
+ * which will be unconditionally freed below.
706
+ */
707
+ free(btf->hdr);
708
+ free(btf->types_data);
709
+ free(btf->strs_data);
710
+ }
711
+ free(btf->raw_data);
712
+ free(btf->raw_data_swapped);
713
+ free(btf->type_offs);
317714 free(btf);
318715 }
319716
320
-struct btf *btf__new(__u8 *data, __u32 size, btf_print_fn_t err_log)
717
+struct btf *btf__new_empty(void)
321718 {
322
- __u32 log_buf_size = 0;
323
- char *log_buf = NULL;
719
+ struct btf *btf;
720
+
721
+ btf = calloc(1, sizeof(*btf));
722
+ if (!btf)
723
+ return ERR_PTR(-ENOMEM);
724
+
725
+ btf->fd = -1;
726
+ btf->ptr_sz = sizeof(void *);
727
+ btf->swapped_endian = false;
728
+
729
+ /* +1 for empty string at offset 0 */
730
+ btf->raw_size = sizeof(struct btf_header) + 1;
731
+ btf->raw_data = calloc(1, btf->raw_size);
732
+ if (!btf->raw_data) {
733
+ free(btf);
734
+ return ERR_PTR(-ENOMEM);
735
+ }
736
+
737
+ btf->hdr = btf->raw_data;
738
+ btf->hdr->hdr_len = sizeof(struct btf_header);
739
+ btf->hdr->magic = BTF_MAGIC;
740
+ btf->hdr->version = BTF_VERSION;
741
+
742
+ btf->types_data = btf->raw_data + btf->hdr->hdr_len;
743
+ btf->strs_data = btf->raw_data + btf->hdr->hdr_len;
744
+ btf->hdr->str_len = 1; /* empty string at offset 0 */
745
+
746
+ return btf;
747
+}
748
+
749
+struct btf *btf__new(const void *data, __u32 size)
750
+{
324751 struct btf *btf;
325752 int err;
326753
....@@ -328,51 +755,30 @@
328755 if (!btf)
329756 return ERR_PTR(-ENOMEM);
330757
331
- btf->fd = -1;
332
-
333
- if (err_log) {
334
- log_buf = malloc(BPF_LOG_BUF_SIZE);
335
- if (!log_buf) {
336
- err = -ENOMEM;
337
- goto done;
338
- }
339
- *log_buf = 0;
340
- log_buf_size = BPF_LOG_BUF_SIZE;
341
- }
342
-
343
- btf->data = malloc(size);
344
- if (!btf->data) {
758
+ btf->raw_data = malloc(size);
759
+ if (!btf->raw_data) {
345760 err = -ENOMEM;
346761 goto done;
347762 }
763
+ memcpy(btf->raw_data, data, size);
764
+ btf->raw_size = size;
348765
349
- memcpy(btf->data, data, size);
350
- btf->data_size = size;
351
-
352
- btf->fd = bpf_load_btf(btf->data, btf->data_size,
353
- log_buf, log_buf_size, false);
354
-
355
- if (btf->fd == -1) {
356
- err = -errno;
357
- elog("Error loading BTF: %s(%d)\n", strerror(errno), errno);
358
- if (log_buf && *log_buf)
359
- elog("%s\n", log_buf);
360
- goto done;
361
- }
362
-
363
- err = btf_parse_hdr(btf, err_log);
766
+ btf->hdr = btf->raw_data;
767
+ err = btf_parse_hdr(btf);
364768 if (err)
365769 goto done;
366770
367
- err = btf_parse_str_sec(btf, err_log);
771
+ btf->strs_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->str_off;
772
+ btf->types_data = btf->raw_data + btf->hdr->hdr_len + btf->hdr->type_off;
773
+
774
+ err = btf_parse_str_sec(btf);
775
+ err = err ?: btf_parse_type_sec(btf);
368776 if (err)
369777 goto done;
370778
371
- err = btf_parse_type_sec(btf, err_log);
779
+ btf->fd = -1;
372780
373781 done:
374
- free(log_buf);
375
-
376782 if (err) {
377783 btf__free(btf);
378784 return ERR_PTR(err);
....@@ -381,15 +787,3685 @@
381787 return btf;
382788 }
383789
790
+struct btf *btf__parse_elf(const char *path, struct btf_ext **btf_ext)
791
+{
792
+ Elf_Data *btf_data = NULL, *btf_ext_data = NULL;
793
+ int err = 0, fd = -1, idx = 0;
794
+ struct btf *btf = NULL;
795
+ Elf_Scn *scn = NULL;
796
+ Elf *elf = NULL;
797
+ GElf_Ehdr ehdr;
798
+
799
+ if (elf_version(EV_CURRENT) == EV_NONE) {
800
+ pr_warn("failed to init libelf for %s\n", path);
801
+ return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
802
+ }
803
+
804
+ fd = open(path, O_RDONLY);
805
+ if (fd < 0) {
806
+ err = -errno;
807
+ pr_warn("failed to open %s: %s\n", path, strerror(errno));
808
+ return ERR_PTR(err);
809
+ }
810
+
811
+ err = -LIBBPF_ERRNO__FORMAT;
812
+
813
+ elf = elf_begin(fd, ELF_C_READ, NULL);
814
+ if (!elf) {
815
+ pr_warn("failed to open %s as ELF file\n", path);
816
+ goto done;
817
+ }
818
+ if (!gelf_getehdr(elf, &ehdr)) {
819
+ pr_warn("failed to get EHDR from %s\n", path);
820
+ goto done;
821
+ }
822
+ if (!elf_rawdata(elf_getscn(elf, ehdr.e_shstrndx), NULL)) {
823
+ pr_warn("failed to get e_shstrndx from %s\n", path);
824
+ goto done;
825
+ }
826
+
827
+ while ((scn = elf_nextscn(elf, scn)) != NULL) {
828
+ GElf_Shdr sh;
829
+ char *name;
830
+
831
+ idx++;
832
+ if (gelf_getshdr(scn, &sh) != &sh) {
833
+ pr_warn("failed to get section(%d) header from %s\n",
834
+ idx, path);
835
+ goto done;
836
+ }
837
+ name = elf_strptr(elf, ehdr.e_shstrndx, sh.sh_name);
838
+ if (!name) {
839
+ pr_warn("failed to get section(%d) name from %s\n",
840
+ idx, path);
841
+ goto done;
842
+ }
843
+ if (strcmp(name, BTF_ELF_SEC) == 0) {
844
+ btf_data = elf_getdata(scn, 0);
845
+ if (!btf_data) {
846
+ pr_warn("failed to get section(%d, %s) data from %s\n",
847
+ idx, name, path);
848
+ goto done;
849
+ }
850
+ continue;
851
+ } else if (btf_ext && strcmp(name, BTF_EXT_ELF_SEC) == 0) {
852
+ btf_ext_data = elf_getdata(scn, 0);
853
+ if (!btf_ext_data) {
854
+ pr_warn("failed to get section(%d, %s) data from %s\n",
855
+ idx, name, path);
856
+ goto done;
857
+ }
858
+ continue;
859
+ }
860
+ }
861
+
862
+ err = 0;
863
+
864
+ if (!btf_data) {
865
+ err = -ENOENT;
866
+ goto done;
867
+ }
868
+ btf = btf__new(btf_data->d_buf, btf_data->d_size);
869
+ if (IS_ERR(btf))
870
+ goto done;
871
+
872
+ switch (gelf_getclass(elf)) {
873
+ case ELFCLASS32:
874
+ btf__set_pointer_size(btf, 4);
875
+ break;
876
+ case ELFCLASS64:
877
+ btf__set_pointer_size(btf, 8);
878
+ break;
879
+ default:
880
+ pr_warn("failed to get ELF class (bitness) for %s\n", path);
881
+ break;
882
+ }
883
+
884
+ if (btf_ext && btf_ext_data) {
885
+ *btf_ext = btf_ext__new(btf_ext_data->d_buf,
886
+ btf_ext_data->d_size);
887
+ if (IS_ERR(*btf_ext))
888
+ goto done;
889
+ } else if (btf_ext) {
890
+ *btf_ext = NULL;
891
+ }
892
+done:
893
+ if (elf)
894
+ elf_end(elf);
895
+ close(fd);
896
+
897
+ if (err)
898
+ return ERR_PTR(err);
899
+ /*
900
+ * btf is always parsed before btf_ext, so no need to clean up
901
+ * btf_ext, if btf loading failed
902
+ */
903
+ if (IS_ERR(btf))
904
+ return btf;
905
+ if (btf_ext && IS_ERR(*btf_ext)) {
906
+ btf__free(btf);
907
+ err = PTR_ERR(*btf_ext);
908
+ return ERR_PTR(err);
909
+ }
910
+ return btf;
911
+}
912
+
913
+struct btf *btf__parse_raw(const char *path)
914
+{
915
+ struct btf *btf = NULL;
916
+ void *data = NULL;
917
+ FILE *f = NULL;
918
+ __u16 magic;
919
+ int err = 0;
920
+ long sz;
921
+
922
+ f = fopen(path, "rb");
923
+ if (!f) {
924
+ err = -errno;
925
+ goto err_out;
926
+ }
927
+
928
+ /* check BTF magic */
929
+ if (fread(&magic, 1, sizeof(magic), f) < sizeof(magic)) {
930
+ err = -EIO;
931
+ goto err_out;
932
+ }
933
+ if (magic != BTF_MAGIC && magic != bswap_16(BTF_MAGIC)) {
934
+ /* definitely not a raw BTF */
935
+ err = -EPROTO;
936
+ goto err_out;
937
+ }
938
+
939
+ /* get file size */
940
+ if (fseek(f, 0, SEEK_END)) {
941
+ err = -errno;
942
+ goto err_out;
943
+ }
944
+ sz = ftell(f);
945
+ if (sz < 0) {
946
+ err = -errno;
947
+ goto err_out;
948
+ }
949
+ /* rewind to the start */
950
+ if (fseek(f, 0, SEEK_SET)) {
951
+ err = -errno;
952
+ goto err_out;
953
+ }
954
+
955
+ /* pre-alloc memory and read all of BTF data */
956
+ data = malloc(sz);
957
+ if (!data) {
958
+ err = -ENOMEM;
959
+ goto err_out;
960
+ }
961
+ if (fread(data, 1, sz, f) < sz) {
962
+ err = -EIO;
963
+ goto err_out;
964
+ }
965
+
966
+ /* finally parse BTF data */
967
+ btf = btf__new(data, sz);
968
+
969
+err_out:
970
+ free(data);
971
+ if (f)
972
+ fclose(f);
973
+ return err ? ERR_PTR(err) : btf;
974
+}
975
+
976
+struct btf *btf__parse(const char *path, struct btf_ext **btf_ext)
977
+{
978
+ struct btf *btf;
979
+
980
+ if (btf_ext)
981
+ *btf_ext = NULL;
982
+
983
+ btf = btf__parse_raw(path);
984
+ if (!IS_ERR(btf) || PTR_ERR(btf) != -EPROTO)
985
+ return btf;
986
+
987
+ return btf__parse_elf(path, btf_ext);
988
+}
989
+
990
+static int compare_vsi_off(const void *_a, const void *_b)
991
+{
992
+ const struct btf_var_secinfo *a = _a;
993
+ const struct btf_var_secinfo *b = _b;
994
+
995
+ return a->offset - b->offset;
996
+}
997
+
998
+static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
999
+ struct btf_type *t)
1000
+{
1001
+ __u32 size = 0, off = 0, i, vars = btf_vlen(t);
1002
+ const char *name = btf__name_by_offset(btf, t->name_off);
1003
+ const struct btf_type *t_var;
1004
+ struct btf_var_secinfo *vsi;
1005
+ const struct btf_var *var;
1006
+ int ret;
1007
+
1008
+ if (!name) {
1009
+ pr_debug("No name found in string section for DATASEC kind.\n");
1010
+ return -ENOENT;
1011
+ }
1012
+
1013
+ /* .extern datasec size and var offsets were set correctly during
1014
+ * extern collection step, so just skip straight to sorting variables
1015
+ */
1016
+ if (t->size)
1017
+ goto sort_vars;
1018
+
1019
+ ret = bpf_object__section_size(obj, name, &size);
1020
+ if (ret || !size || (t->size && t->size != size)) {
1021
+ pr_debug("Invalid size for section %s: %u bytes\n", name, size);
1022
+ return -ENOENT;
1023
+ }
1024
+
1025
+ t->size = size;
1026
+
1027
+ for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
1028
+ t_var = btf__type_by_id(btf, vsi->type);
1029
+ var = btf_var(t_var);
1030
+
1031
+ if (!btf_is_var(t_var)) {
1032
+ pr_debug("Non-VAR type seen in section %s\n", name);
1033
+ return -EINVAL;
1034
+ }
1035
+
1036
+ if (var->linkage == BTF_VAR_STATIC)
1037
+ continue;
1038
+
1039
+ name = btf__name_by_offset(btf, t_var->name_off);
1040
+ if (!name) {
1041
+ pr_debug("No name found in string section for VAR kind\n");
1042
+ return -ENOENT;
1043
+ }
1044
+
1045
+ ret = bpf_object__variable_offset(obj, name, &off);
1046
+ if (ret) {
1047
+ pr_debug("No offset found in symbol table for VAR %s\n",
1048
+ name);
1049
+ return -ENOENT;
1050
+ }
1051
+
1052
+ vsi->offset = off;
1053
+ }
1054
+
1055
+sort_vars:
1056
+ qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
1057
+ return 0;
1058
+}
1059
+
1060
+int btf__finalize_data(struct bpf_object *obj, struct btf *btf)
1061
+{
1062
+ int err = 0;
1063
+ __u32 i;
1064
+
1065
+ for (i = 1; i <= btf->nr_types; i++) {
1066
+ struct btf_type *t = btf_type_by_id(btf, i);
1067
+
1068
+ /* Loader needs to fix up some of the things compiler
1069
+ * couldn't get its hands on while emitting BTF. This
1070
+ * is section size and global variable offset. We use
1071
+ * the info from the ELF itself for this purpose.
1072
+ */
1073
+ if (btf_is_datasec(t)) {
1074
+ err = btf_fixup_datasec(obj, btf, t);
1075
+ if (err)
1076
+ break;
1077
+ }
1078
+ }
1079
+
1080
+ return err;
1081
+}
1082
+
1083
+static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian);
1084
+
1085
+int btf__load(struct btf *btf)
1086
+{
1087
+ __u32 log_buf_size = 0, raw_size;
1088
+ char *log_buf = NULL;
1089
+ void *raw_data;
1090
+ int err = 0;
1091
+
1092
+ if (btf->fd >= 0)
1093
+ return -EEXIST;
1094
+
1095
+retry_load:
1096
+ if (log_buf_size) {
1097
+ log_buf = malloc(log_buf_size);
1098
+ if (!log_buf)
1099
+ return -ENOMEM;
1100
+
1101
+ *log_buf = 0;
1102
+ }
1103
+
1104
+ raw_data = btf_get_raw_data(btf, &raw_size, false);
1105
+ if (!raw_data) {
1106
+ err = -ENOMEM;
1107
+ goto done;
1108
+ }
1109
+ /* cache native raw data representation */
1110
+ btf->raw_size = raw_size;
1111
+ btf->raw_data = raw_data;
1112
+
1113
+ btf->fd = bpf_load_btf(raw_data, raw_size, log_buf, log_buf_size, false);
1114
+ if (btf->fd < 0) {
1115
+ if (!log_buf || errno == ENOSPC) {
1116
+ log_buf_size = max((__u32)BPF_LOG_BUF_SIZE,
1117
+ log_buf_size << 1);
1118
+ free(log_buf);
1119
+ goto retry_load;
1120
+ }
1121
+
1122
+ err = -errno;
1123
+ pr_warn("Error loading BTF: %s(%d)\n", strerror(errno), errno);
1124
+ if (*log_buf)
1125
+ pr_warn("%s\n", log_buf);
1126
+ goto done;
1127
+ }
1128
+
1129
+done:
1130
+ free(log_buf);
1131
+ return err;
1132
+}
1133
+
3841134 int btf__fd(const struct btf *btf)
3851135 {
3861136 return btf->fd;
3871137 }
3881138
389
-const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
1139
+void btf__set_fd(struct btf *btf, int fd)
1140
+{
1141
+ btf->fd = fd;
1142
+}
1143
+
1144
+static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian)
1145
+{
1146
+ struct btf_header *hdr = btf->hdr;
1147
+ struct btf_type *t;
1148
+ void *data, *p;
1149
+ __u32 data_sz;
1150
+ int i;
1151
+
1152
+ data = swap_endian ? btf->raw_data_swapped : btf->raw_data;
1153
+ if (data) {
1154
+ *size = btf->raw_size;
1155
+ return data;
1156
+ }
1157
+
1158
+ data_sz = hdr->hdr_len + hdr->type_len + hdr->str_len;
1159
+ data = calloc(1, data_sz);
1160
+ if (!data)
1161
+ return NULL;
1162
+ p = data;
1163
+
1164
+ memcpy(p, hdr, hdr->hdr_len);
1165
+ if (swap_endian)
1166
+ btf_bswap_hdr(p);
1167
+ p += hdr->hdr_len;
1168
+
1169
+ memcpy(p, btf->types_data, hdr->type_len);
1170
+ if (swap_endian) {
1171
+ for (i = 1; i <= btf->nr_types; i++) {
1172
+ t = p + btf->type_offs[i];
1173
+ /* btf_bswap_type_rest() relies on native t->info, so
1174
+ * we swap base type info after we swapped all the
1175
+ * additional information
1176
+ */
1177
+ if (btf_bswap_type_rest(t))
1178
+ goto err_out;
1179
+ btf_bswap_type_base(t);
1180
+ }
1181
+ }
1182
+ p += hdr->type_len;
1183
+
1184
+ memcpy(p, btf->strs_data, hdr->str_len);
1185
+ p += hdr->str_len;
1186
+
1187
+ *size = data_sz;
1188
+ return data;
1189
+err_out:
1190
+ free(data);
1191
+ return NULL;
1192
+}
1193
+
1194
+const void *btf__get_raw_data(const struct btf *btf_ro, __u32 *size)
1195
+{
1196
+ struct btf *btf = (struct btf *)btf_ro;
1197
+ __u32 data_sz;
1198
+ void *data;
1199
+
1200
+ data = btf_get_raw_data(btf, &data_sz, btf->swapped_endian);
1201
+ if (!data)
1202
+ return NULL;
1203
+
1204
+ btf->raw_size = data_sz;
1205
+ if (btf->swapped_endian)
1206
+ btf->raw_data_swapped = data;
1207
+ else
1208
+ btf->raw_data = data;
1209
+ *size = data_sz;
1210
+ return data;
1211
+}
1212
+
1213
+const char *btf__str_by_offset(const struct btf *btf, __u32 offset)
3901214 {
3911215 if (offset < btf->hdr->str_len)
392
- return &btf->strings[offset];
1216
+ return btf->strs_data + offset;
3931217 else
3941218 return NULL;
3951219 }
1220
+
1221
+const char *btf__name_by_offset(const struct btf *btf, __u32 offset)
1222
+{
1223
+ return btf__str_by_offset(btf, offset);
1224
+}
1225
+
1226
+int btf__get_from_id(__u32 id, struct btf **btf)
1227
+{
1228
+ struct bpf_btf_info btf_info = { 0 };
1229
+ __u32 len = sizeof(btf_info);
1230
+ __u32 last_size;
1231
+ int btf_fd;
1232
+ void *ptr;
1233
+ int err;
1234
+
1235
+ err = 0;
1236
+ *btf = NULL;
1237
+ btf_fd = bpf_btf_get_fd_by_id(id);
1238
+ if (btf_fd < 0)
1239
+ return 0;
1240
+
1241
+ /* we won't know btf_size until we call bpf_obj_get_info_by_fd(). so
1242
+ * let's start with a sane default - 4KiB here - and resize it only if
1243
+ * bpf_obj_get_info_by_fd() needs a bigger buffer.
1244
+ */
1245
+ btf_info.btf_size = 4096;
1246
+ last_size = btf_info.btf_size;
1247
+ ptr = malloc(last_size);
1248
+ if (!ptr) {
1249
+ err = -ENOMEM;
1250
+ goto exit_free;
1251
+ }
1252
+
1253
+ memset(ptr, 0, last_size);
1254
+ btf_info.btf = ptr_to_u64(ptr);
1255
+ err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1256
+
1257
+ if (!err && btf_info.btf_size > last_size) {
1258
+ void *temp_ptr;
1259
+
1260
+ last_size = btf_info.btf_size;
1261
+ temp_ptr = realloc(ptr, last_size);
1262
+ if (!temp_ptr) {
1263
+ err = -ENOMEM;
1264
+ goto exit_free;
1265
+ }
1266
+ ptr = temp_ptr;
1267
+ memset(ptr, 0, last_size);
1268
+ btf_info.btf = ptr_to_u64(ptr);
1269
+ err = bpf_obj_get_info_by_fd(btf_fd, &btf_info, &len);
1270
+ }
1271
+
1272
+ if (err || btf_info.btf_size > last_size) {
1273
+ err = errno;
1274
+ goto exit_free;
1275
+ }
1276
+
1277
+ *btf = btf__new((__u8 *)(long)btf_info.btf, btf_info.btf_size);
1278
+ if (IS_ERR(*btf)) {
1279
+ err = PTR_ERR(*btf);
1280
+ *btf = NULL;
1281
+ }
1282
+
1283
+exit_free:
1284
+ close(btf_fd);
1285
+ free(ptr);
1286
+
1287
+ return err;
1288
+}
1289
+
1290
+int btf__get_map_kv_tids(const struct btf *btf, const char *map_name,
1291
+ __u32 expected_key_size, __u32 expected_value_size,
1292
+ __u32 *key_type_id, __u32 *value_type_id)
1293
+{
1294
+ const struct btf_type *container_type;
1295
+ const struct btf_member *key, *value;
1296
+ const size_t max_name = 256;
1297
+ char container_name[max_name];
1298
+ __s64 key_size, value_size;
1299
+ __s32 container_id;
1300
+
1301
+ if (snprintf(container_name, max_name, "____btf_map_%s", map_name) ==
1302
+ max_name) {
1303
+ pr_warn("map:%s length of '____btf_map_%s' is too long\n",
1304
+ map_name, map_name);
1305
+ return -EINVAL;
1306
+ }
1307
+
1308
+ container_id = btf__find_by_name(btf, container_name);
1309
+ if (container_id < 0) {
1310
+ pr_debug("map:%s container_name:%s cannot be found in BTF. Missing BPF_ANNOTATE_KV_PAIR?\n",
1311
+ map_name, container_name);
1312
+ return container_id;
1313
+ }
1314
+
1315
+ container_type = btf__type_by_id(btf, container_id);
1316
+ if (!container_type) {
1317
+ pr_warn("map:%s cannot find BTF type for container_id:%u\n",
1318
+ map_name, container_id);
1319
+ return -EINVAL;
1320
+ }
1321
+
1322
+ if (!btf_is_struct(container_type) || btf_vlen(container_type) < 2) {
1323
+ pr_warn("map:%s container_name:%s is an invalid container struct\n",
1324
+ map_name, container_name);
1325
+ return -EINVAL;
1326
+ }
1327
+
1328
+ key = btf_members(container_type);
1329
+ value = key + 1;
1330
+
1331
+ key_size = btf__resolve_size(btf, key->type);
1332
+ if (key_size < 0) {
1333
+ pr_warn("map:%s invalid BTF key_type_size\n", map_name);
1334
+ return key_size;
1335
+ }
1336
+
1337
+ if (expected_key_size != key_size) {
1338
+ pr_warn("map:%s btf_key_type_size:%u != map_def_key_size:%u\n",
1339
+ map_name, (__u32)key_size, expected_key_size);
1340
+ return -EINVAL;
1341
+ }
1342
+
1343
+ value_size = btf__resolve_size(btf, value->type);
1344
+ if (value_size < 0) {
1345
+ pr_warn("map:%s invalid BTF value_type_size\n", map_name);
1346
+ return value_size;
1347
+ }
1348
+
1349
+ if (expected_value_size != value_size) {
1350
+ pr_warn("map:%s btf_value_type_size:%u != map_def_value_size:%u\n",
1351
+ map_name, (__u32)value_size, expected_value_size);
1352
+ return -EINVAL;
1353
+ }
1354
+
1355
+ *key_type_id = key->type;
1356
+ *value_type_id = value->type;
1357
+
1358
+ return 0;
1359
+}
1360
+
1361
+static size_t strs_hash_fn(const void *key, void *ctx)
1362
+{
1363
+ struct btf *btf = ctx;
1364
+ const char *str = btf->strs_data + (long)key;
1365
+
1366
+ return str_hash(str);
1367
+}
1368
+
1369
+static bool strs_hash_equal_fn(const void *key1, const void *key2, void *ctx)
1370
+{
1371
+ struct btf *btf = ctx;
1372
+ const char *str1 = btf->strs_data + (long)key1;
1373
+ const char *str2 = btf->strs_data + (long)key2;
1374
+
1375
+ return strcmp(str1, str2) == 0;
1376
+}
1377
+
1378
+static void btf_invalidate_raw_data(struct btf *btf)
1379
+{
1380
+ if (btf->raw_data) {
1381
+ free(btf->raw_data);
1382
+ btf->raw_data = NULL;
1383
+ }
1384
+ if (btf->raw_data_swapped) {
1385
+ free(btf->raw_data_swapped);
1386
+ btf->raw_data_swapped = NULL;
1387
+ }
1388
+}
1389
+
1390
+/* Ensure BTF is ready to be modified (by splitting into a three memory
1391
+ * regions for header, types, and strings). Also invalidate cached
1392
+ * raw_data, if any.
1393
+ */
1394
+static int btf_ensure_modifiable(struct btf *btf)
1395
+{
1396
+ void *hdr, *types, *strs, *strs_end, *s;
1397
+ struct hashmap *hash = NULL;
1398
+ long off;
1399
+ int err;
1400
+
1401
+ if (btf_is_modifiable(btf)) {
1402
+ /* any BTF modification invalidates raw_data */
1403
+ btf_invalidate_raw_data(btf);
1404
+ return 0;
1405
+ }
1406
+
1407
+ /* split raw data into three memory regions */
1408
+ hdr = malloc(btf->hdr->hdr_len);
1409
+ types = malloc(btf->hdr->type_len);
1410
+ strs = malloc(btf->hdr->str_len);
1411
+ if (!hdr || !types || !strs)
1412
+ goto err_out;
1413
+
1414
+ memcpy(hdr, btf->hdr, btf->hdr->hdr_len);
1415
+ memcpy(types, btf->types_data, btf->hdr->type_len);
1416
+ memcpy(strs, btf->strs_data, btf->hdr->str_len);
1417
+
1418
+ /* build lookup index for all strings */
1419
+ hash = hashmap__new(strs_hash_fn, strs_hash_equal_fn, btf);
1420
+ if (IS_ERR(hash)) {
1421
+ err = PTR_ERR(hash);
1422
+ hash = NULL;
1423
+ goto err_out;
1424
+ }
1425
+
1426
+ strs_end = strs + btf->hdr->str_len;
1427
+ for (off = 0, s = strs; s < strs_end; off += strlen(s) + 1, s = strs + off) {
1428
+ /* hashmap__add() returns EEXIST if string with the same
1429
+ * content already is in the hash map
1430
+ */
1431
+ err = hashmap__add(hash, (void *)off, (void *)off);
1432
+ if (err == -EEXIST)
1433
+ continue; /* duplicate */
1434
+ if (err)
1435
+ goto err_out;
1436
+ }
1437
+
1438
+ /* only when everything was successful, update internal state */
1439
+ btf->hdr = hdr;
1440
+ btf->types_data = types;
1441
+ btf->types_data_cap = btf->hdr->type_len;
1442
+ btf->strs_data = strs;
1443
+ btf->strs_data_cap = btf->hdr->str_len;
1444
+ btf->strs_hash = hash;
1445
+ /* if BTF was created from scratch, all strings are guaranteed to be
1446
+ * unique and deduplicated
1447
+ */
1448
+ btf->strs_deduped = btf->hdr->str_len <= 1;
1449
+
1450
+ /* invalidate raw_data representation */
1451
+ btf_invalidate_raw_data(btf);
1452
+
1453
+ return 0;
1454
+
1455
+err_out:
1456
+ hashmap__free(hash);
1457
+ free(hdr);
1458
+ free(types);
1459
+ free(strs);
1460
+ return -ENOMEM;
1461
+}
1462
+
1463
+static void *btf_add_str_mem(struct btf *btf, size_t add_sz)
1464
+{
1465
+ return btf_add_mem(&btf->strs_data, &btf->strs_data_cap, 1,
1466
+ btf->hdr->str_len, BTF_MAX_STR_OFFSET, add_sz);
1467
+}
1468
+
1469
+/* Find an offset in BTF string section that corresponds to a given string *s*.
1470
+ * Returns:
1471
+ * - >0 offset into string section, if string is found;
1472
+ * - -ENOENT, if string is not in the string section;
1473
+ * - <0, on any other error.
1474
+ */
1475
+int btf__find_str(struct btf *btf, const char *s)
1476
+{
1477
+ long old_off, new_off, len;
1478
+ void *p;
1479
+
1480
+ /* BTF needs to be in a modifiable state to build string lookup index */
1481
+ if (btf_ensure_modifiable(btf))
1482
+ return -ENOMEM;
1483
+
1484
+ /* see btf__add_str() for why we do this */
1485
+ len = strlen(s) + 1;
1486
+ p = btf_add_str_mem(btf, len);
1487
+ if (!p)
1488
+ return -ENOMEM;
1489
+
1490
+ new_off = btf->hdr->str_len;
1491
+ memcpy(p, s, len);
1492
+
1493
+ if (hashmap__find(btf->strs_hash, (void *)new_off, (void **)&old_off))
1494
+ return old_off;
1495
+
1496
+ return -ENOENT;
1497
+}
1498
+
1499
+/* Add a string s to the BTF string section.
1500
+ * Returns:
1501
+ * - > 0 offset into string section, on success;
1502
+ * - < 0, on error.
1503
+ */
1504
+int btf__add_str(struct btf *btf, const char *s)
1505
+{
1506
+ long old_off, new_off, len;
1507
+ void *p;
1508
+ int err;
1509
+
1510
+ if (btf_ensure_modifiable(btf))
1511
+ return -ENOMEM;
1512
+
1513
+ /* Hashmap keys are always offsets within btf->strs_data, so to even
1514
+ * look up some string from the "outside", we need to first append it
1515
+ * at the end, so that it can be addressed with an offset. Luckily,
1516
+ * until btf->hdr->str_len is incremented, that string is just a piece
1517
+ * of garbage for the rest of BTF code, so no harm, no foul. On the
1518
+ * other hand, if the string is unique, it's already appended and
1519
+ * ready to be used, only a simple btf->hdr->str_len increment away.
1520
+ */
1521
+ len = strlen(s) + 1;
1522
+ p = btf_add_str_mem(btf, len);
1523
+ if (!p)
1524
+ return -ENOMEM;
1525
+
1526
+ new_off = btf->hdr->str_len;
1527
+ memcpy(p, s, len);
1528
+
1529
+ /* Now attempt to add the string, but only if the string with the same
1530
+ * contents doesn't exist already (HASHMAP_ADD strategy). If such
1531
+ * string exists, we'll get its offset in old_off (that's old_key).
1532
+ */
1533
+ err = hashmap__insert(btf->strs_hash, (void *)new_off, (void *)new_off,
1534
+ HASHMAP_ADD, (const void **)&old_off, NULL);
1535
+ if (err == -EEXIST)
1536
+ return old_off; /* duplicated string, return existing offset */
1537
+ if (err)
1538
+ return err;
1539
+
1540
+ btf->hdr->str_len += len; /* new unique string, adjust data length */
1541
+ return new_off;
1542
+}
1543
+
1544
+static void *btf_add_type_mem(struct btf *btf, size_t add_sz)
1545
+{
1546
+ return btf_add_mem(&btf->types_data, &btf->types_data_cap, 1,
1547
+ btf->hdr->type_len, UINT_MAX, add_sz);
1548
+}
1549
+
1550
+static __u32 btf_type_info(int kind, int vlen, int kflag)
1551
+{
1552
+ return (kflag << 31) | (kind << 24) | vlen;
1553
+}
1554
+
1555
+static void btf_type_inc_vlen(struct btf_type *t)
1556
+{
1557
+ t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, btf_kflag(t));
1558
+}
1559
+
1560
+/*
1561
+ * Append new BTF_KIND_INT type with:
1562
+ * - *name* - non-empty, non-NULL type name;
1563
+ * - *sz* - power-of-2 (1, 2, 4, ..) size of the type, in bytes;
1564
+ * - encoding is a combination of BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL.
1565
+ * Returns:
1566
+ * - >0, type ID of newly added BTF type;
1567
+ * - <0, on error.
1568
+ */
1569
+int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding)
1570
+{
1571
+ struct btf_type *t;
1572
+ int sz, err, name_off;
1573
+
1574
+ /* non-empty name */
1575
+ if (!name || !name[0])
1576
+ return -EINVAL;
1577
+ /* byte_sz must be power of 2 */
1578
+ if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
1579
+ return -EINVAL;
1580
+ if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
1581
+ return -EINVAL;
1582
+
1583
+ /* deconstruct BTF, if necessary, and invalidate raw_data */
1584
+ if (btf_ensure_modifiable(btf))
1585
+ return -ENOMEM;
1586
+
1587
+ sz = sizeof(struct btf_type) + sizeof(int);
1588
+ t = btf_add_type_mem(btf, sz);
1589
+ if (!t)
1590
+ return -ENOMEM;
1591
+
1592
+ /* if something goes wrong later, we might end up with an extra string,
1593
+ * but that shouldn't be a problem, because BTF can't be constructed
1594
+ * completely anyway and will most probably be just discarded
1595
+ */
1596
+ name_off = btf__add_str(btf, name);
1597
+ if (name_off < 0)
1598
+ return name_off;
1599
+
1600
+ t->name_off = name_off;
1601
+ t->info = btf_type_info(BTF_KIND_INT, 0, 0);
1602
+ t->size = byte_sz;
1603
+ /* set INT info, we don't allow setting legacy bit offset/size */
1604
+ *(__u32 *)(t + 1) = (encoding << 24) | (byte_sz * 8);
1605
+
1606
+ err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1607
+ if (err)
1608
+ return err;
1609
+
1610
+ btf->hdr->type_len += sz;
1611
+ btf->hdr->str_off += sz;
1612
+ btf->nr_types++;
1613
+ return btf->nr_types;
1614
+}
1615
+
1616
+/* it's completely legal to append BTF types with type IDs pointing forward to
1617
+ * types that haven't been appended yet, so we only make sure that id looks
1618
+ * sane, we can't guarantee that ID will always be valid
1619
+ */
1620
+static int validate_type_id(int id)
1621
+{
1622
+ if (id < 0 || id > BTF_MAX_NR_TYPES)
1623
+ return -EINVAL;
1624
+ return 0;
1625
+}
1626
+
1627
+/* generic append function for PTR, TYPEDEF, CONST/VOLATILE/RESTRICT */
1628
+static int btf_add_ref_kind(struct btf *btf, int kind, const char *name, int ref_type_id)
1629
+{
1630
+ struct btf_type *t;
1631
+ int sz, name_off = 0, err;
1632
+
1633
+ if (validate_type_id(ref_type_id))
1634
+ return -EINVAL;
1635
+
1636
+ if (btf_ensure_modifiable(btf))
1637
+ return -ENOMEM;
1638
+
1639
+ sz = sizeof(struct btf_type);
1640
+ t = btf_add_type_mem(btf, sz);
1641
+ if (!t)
1642
+ return -ENOMEM;
1643
+
1644
+ if (name && name[0]) {
1645
+ name_off = btf__add_str(btf, name);
1646
+ if (name_off < 0)
1647
+ return name_off;
1648
+ }
1649
+
1650
+ t->name_off = name_off;
1651
+ t->info = btf_type_info(kind, 0, 0);
1652
+ t->type = ref_type_id;
1653
+
1654
+ err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1655
+ if (err)
1656
+ return err;
1657
+
1658
+ btf->hdr->type_len += sz;
1659
+ btf->hdr->str_off += sz;
1660
+ btf->nr_types++;
1661
+ return btf->nr_types;
1662
+}
1663
+
1664
+/*
1665
+ * Append new BTF_KIND_PTR type with:
1666
+ * - *ref_type_id* - referenced type ID, it might not exist yet;
1667
+ * Returns:
1668
+ * - >0, type ID of newly added BTF type;
1669
+ * - <0, on error.
1670
+ */
1671
+int btf__add_ptr(struct btf *btf, int ref_type_id)
1672
+{
1673
+ return btf_add_ref_kind(btf, BTF_KIND_PTR, NULL, ref_type_id);
1674
+}
1675
+
1676
+/*
1677
+ * Append new BTF_KIND_ARRAY type with:
1678
+ * - *index_type_id* - type ID of the type describing array index;
1679
+ * - *elem_type_id* - type ID of the type describing array element;
1680
+ * - *nr_elems* - the size of the array;
1681
+ * Returns:
1682
+ * - >0, type ID of newly added BTF type;
1683
+ * - <0, on error.
1684
+ */
1685
+int btf__add_array(struct btf *btf, int index_type_id, int elem_type_id, __u32 nr_elems)
1686
+{
1687
+ struct btf_type *t;
1688
+ struct btf_array *a;
1689
+ int sz, err;
1690
+
1691
+ if (validate_type_id(index_type_id) || validate_type_id(elem_type_id))
1692
+ return -EINVAL;
1693
+
1694
+ if (btf_ensure_modifiable(btf))
1695
+ return -ENOMEM;
1696
+
1697
+ sz = sizeof(struct btf_type) + sizeof(struct btf_array);
1698
+ t = btf_add_type_mem(btf, sz);
1699
+ if (!t)
1700
+ return -ENOMEM;
1701
+
1702
+ t->name_off = 0;
1703
+ t->info = btf_type_info(BTF_KIND_ARRAY, 0, 0);
1704
+ t->size = 0;
1705
+
1706
+ a = btf_array(t);
1707
+ a->type = elem_type_id;
1708
+ a->index_type = index_type_id;
1709
+ a->nelems = nr_elems;
1710
+
1711
+ err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1712
+ if (err)
1713
+ return err;
1714
+
1715
+ btf->hdr->type_len += sz;
1716
+ btf->hdr->str_off += sz;
1717
+ btf->nr_types++;
1718
+ return btf->nr_types;
1719
+}
1720
+
1721
+/* generic STRUCT/UNION append function */
1722
+static int btf_add_composite(struct btf *btf, int kind, const char *name, __u32 bytes_sz)
1723
+{
1724
+ struct btf_type *t;
1725
+ int sz, err, name_off = 0;
1726
+
1727
+ if (btf_ensure_modifiable(btf))
1728
+ return -ENOMEM;
1729
+
1730
+ sz = sizeof(struct btf_type);
1731
+ t = btf_add_type_mem(btf, sz);
1732
+ if (!t)
1733
+ return -ENOMEM;
1734
+
1735
+ if (name && name[0]) {
1736
+ name_off = btf__add_str(btf, name);
1737
+ if (name_off < 0)
1738
+ return name_off;
1739
+ }
1740
+
1741
+ /* start out with vlen=0 and no kflag; this will be adjusted when
1742
+ * adding each member
1743
+ */
1744
+ t->name_off = name_off;
1745
+ t->info = btf_type_info(kind, 0, 0);
1746
+ t->size = bytes_sz;
1747
+
1748
+ err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1749
+ if (err)
1750
+ return err;
1751
+
1752
+ btf->hdr->type_len += sz;
1753
+ btf->hdr->str_off += sz;
1754
+ btf->nr_types++;
1755
+ return btf->nr_types;
1756
+}
1757
+
1758
+/*
1759
+ * Append new BTF_KIND_STRUCT type with:
1760
+ * - *name* - name of the struct, can be NULL or empty for anonymous structs;
1761
+ * - *byte_sz* - size of the struct, in bytes;
1762
+ *
1763
+ * Struct initially has no fields in it. Fields can be added by
1764
+ * btf__add_field() right after btf__add_struct() succeeds.
1765
+ *
1766
+ * Returns:
1767
+ * - >0, type ID of newly added BTF type;
1768
+ * - <0, on error.
1769
+ */
1770
+int btf__add_struct(struct btf *btf, const char *name, __u32 byte_sz)
1771
+{
1772
+ return btf_add_composite(btf, BTF_KIND_STRUCT, name, byte_sz);
1773
+}
1774
+
1775
+/*
1776
+ * Append new BTF_KIND_UNION type with:
1777
+ * - *name* - name of the union, can be NULL or empty for anonymous union;
1778
+ * - *byte_sz* - size of the union, in bytes;
1779
+ *
1780
+ * Union initially has no fields in it. Fields can be added by
1781
+ * btf__add_field() right after btf__add_union() succeeds. All fields
1782
+ * should have *bit_offset* of 0.
1783
+ *
1784
+ * Returns:
1785
+ * - >0, type ID of newly added BTF type;
1786
+ * - <0, on error.
1787
+ */
1788
+int btf__add_union(struct btf *btf, const char *name, __u32 byte_sz)
1789
+{
1790
+ return btf_add_composite(btf, BTF_KIND_UNION, name, byte_sz);
1791
+}
1792
+
1793
+/*
1794
+ * Append new field for the current STRUCT/UNION type with:
1795
+ * - *name* - name of the field, can be NULL or empty for anonymous field;
1796
+ * - *type_id* - type ID for the type describing field type;
1797
+ * - *bit_offset* - bit offset of the start of the field within struct/union;
1798
+ * - *bit_size* - bit size of a bitfield, 0 for non-bitfield fields;
1799
+ * Returns:
1800
+ * - 0, on success;
1801
+ * - <0, on error.
1802
+ */
1803
+int btf__add_field(struct btf *btf, const char *name, int type_id,
1804
+ __u32 bit_offset, __u32 bit_size)
1805
+{
1806
+ struct btf_type *t;
1807
+ struct btf_member *m;
1808
+ bool is_bitfield;
1809
+ int sz, name_off = 0;
1810
+
1811
+ /* last type should be union/struct */
1812
+ if (btf->nr_types == 0)
1813
+ return -EINVAL;
1814
+ t = btf_type_by_id(btf, btf->nr_types);
1815
+ if (!btf_is_composite(t))
1816
+ return -EINVAL;
1817
+
1818
+ if (validate_type_id(type_id))
1819
+ return -EINVAL;
1820
+ /* best-effort bit field offset/size enforcement */
1821
+ is_bitfield = bit_size || (bit_offset % 8 != 0);
1822
+ if (is_bitfield && (bit_size == 0 || bit_size > 255 || bit_offset > 0xffffff))
1823
+ return -EINVAL;
1824
+
1825
+ /* only offset 0 is allowed for unions */
1826
+ if (btf_is_union(t) && bit_offset)
1827
+ return -EINVAL;
1828
+
1829
+ /* decompose and invalidate raw data */
1830
+ if (btf_ensure_modifiable(btf))
1831
+ return -ENOMEM;
1832
+
1833
+ sz = sizeof(struct btf_member);
1834
+ m = btf_add_type_mem(btf, sz);
1835
+ if (!m)
1836
+ return -ENOMEM;
1837
+
1838
+ if (name && name[0]) {
1839
+ name_off = btf__add_str(btf, name);
1840
+ if (name_off < 0)
1841
+ return name_off;
1842
+ }
1843
+
1844
+ m->name_off = name_off;
1845
+ m->type = type_id;
1846
+ m->offset = bit_offset | (bit_size << 24);
1847
+
1848
+ /* btf_add_type_mem can invalidate t pointer */
1849
+ t = btf_type_by_id(btf, btf->nr_types);
1850
+ /* update parent type's vlen and kflag */
1851
+ t->info = btf_type_info(btf_kind(t), btf_vlen(t) + 1, is_bitfield || btf_kflag(t));
1852
+
1853
+ btf->hdr->type_len += sz;
1854
+ btf->hdr->str_off += sz;
1855
+ return 0;
1856
+}
1857
+
1858
+/*
1859
+ * Append new BTF_KIND_ENUM type with:
1860
+ * - *name* - name of the enum, can be NULL or empty for anonymous enums;
1861
+ * - *byte_sz* - size of the enum, in bytes.
1862
+ *
1863
+ * Enum initially has no enum values in it (and corresponds to enum forward
1864
+ * declaration). Enumerator values can be added by btf__add_enum_value()
1865
+ * immediately after btf__add_enum() succeeds.
1866
+ *
1867
+ * Returns:
1868
+ * - >0, type ID of newly added BTF type;
1869
+ * - <0, on error.
1870
+ */
1871
+int btf__add_enum(struct btf *btf, const char *name, __u32 byte_sz)
1872
+{
1873
+ struct btf_type *t;
1874
+ int sz, err, name_off = 0;
1875
+
1876
+ /* byte_sz must be power of 2 */
1877
+ if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 8)
1878
+ return -EINVAL;
1879
+
1880
+ if (btf_ensure_modifiable(btf))
1881
+ return -ENOMEM;
1882
+
1883
+ sz = sizeof(struct btf_type);
1884
+ t = btf_add_type_mem(btf, sz);
1885
+ if (!t)
1886
+ return -ENOMEM;
1887
+
1888
+ if (name && name[0]) {
1889
+ name_off = btf__add_str(btf, name);
1890
+ if (name_off < 0)
1891
+ return name_off;
1892
+ }
1893
+
1894
+ /* start out with vlen=0; it will be adjusted when adding enum values */
1895
+ t->name_off = name_off;
1896
+ t->info = btf_type_info(BTF_KIND_ENUM, 0, 0);
1897
+ t->size = byte_sz;
1898
+
1899
+ err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
1900
+ if (err)
1901
+ return err;
1902
+
1903
+ btf->hdr->type_len += sz;
1904
+ btf->hdr->str_off += sz;
1905
+ btf->nr_types++;
1906
+ return btf->nr_types;
1907
+}
1908
+
1909
+/*
1910
+ * Append new enum value for the current ENUM type with:
1911
+ * - *name* - name of the enumerator value, can't be NULL or empty;
1912
+ * - *value* - integer value corresponding to enum value *name*;
1913
+ * Returns:
1914
+ * - 0, on success;
1915
+ * - <0, on error.
1916
+ */
1917
+int btf__add_enum_value(struct btf *btf, const char *name, __s64 value)
1918
+{
1919
+ struct btf_type *t;
1920
+ struct btf_enum *v;
1921
+ int sz, name_off;
1922
+
1923
+ /* last type should be BTF_KIND_ENUM */
1924
+ if (btf->nr_types == 0)
1925
+ return -EINVAL;
1926
+ t = btf_type_by_id(btf, btf->nr_types);
1927
+ if (!btf_is_enum(t))
1928
+ return -EINVAL;
1929
+
1930
+ /* non-empty name */
1931
+ if (!name || !name[0])
1932
+ return -EINVAL;
1933
+ if (value < INT_MIN || value > UINT_MAX)
1934
+ return -E2BIG;
1935
+
1936
+ /* decompose and invalidate raw data */
1937
+ if (btf_ensure_modifiable(btf))
1938
+ return -ENOMEM;
1939
+
1940
+ sz = sizeof(struct btf_enum);
1941
+ v = btf_add_type_mem(btf, sz);
1942
+ if (!v)
1943
+ return -ENOMEM;
1944
+
1945
+ name_off = btf__add_str(btf, name);
1946
+ if (name_off < 0)
1947
+ return name_off;
1948
+
1949
+ v->name_off = name_off;
1950
+ v->val = value;
1951
+
1952
+ /* update parent type's vlen */
1953
+ t = btf_type_by_id(btf, btf->nr_types);
1954
+ btf_type_inc_vlen(t);
1955
+
1956
+ btf->hdr->type_len += sz;
1957
+ btf->hdr->str_off += sz;
1958
+ return 0;
1959
+}
1960
+
1961
+/*
1962
+ * Append new BTF_KIND_FWD type with:
1963
+ * - *name*, non-empty/non-NULL name;
1964
+ * - *fwd_kind*, kind of forward declaration, one of BTF_FWD_STRUCT,
1965
+ * BTF_FWD_UNION, or BTF_FWD_ENUM;
1966
+ * Returns:
1967
+ * - >0, type ID of newly added BTF type;
1968
+ * - <0, on error.
1969
+ */
1970
+int btf__add_fwd(struct btf *btf, const char *name, enum btf_fwd_kind fwd_kind)
1971
+{
1972
+ if (!name || !name[0])
1973
+ return -EINVAL;
1974
+
1975
+ switch (fwd_kind) {
1976
+ case BTF_FWD_STRUCT:
1977
+ case BTF_FWD_UNION: {
1978
+ struct btf_type *t;
1979
+ int id;
1980
+
1981
+ id = btf_add_ref_kind(btf, BTF_KIND_FWD, name, 0);
1982
+ if (id <= 0)
1983
+ return id;
1984
+ t = btf_type_by_id(btf, id);
1985
+ t->info = btf_type_info(BTF_KIND_FWD, 0, fwd_kind == BTF_FWD_UNION);
1986
+ return id;
1987
+ }
1988
+ case BTF_FWD_ENUM:
1989
+ /* enum forward in BTF currently is just an enum with no enum
1990
+ * values; we also assume a standard 4-byte size for it
1991
+ */
1992
+ return btf__add_enum(btf, name, sizeof(int));
1993
+ default:
1994
+ return -EINVAL;
1995
+ }
1996
+}
1997
+
1998
+/*
1999
+ * Append new BTF_KING_TYPEDEF type with:
2000
+ * - *name*, non-empty/non-NULL name;
2001
+ * - *ref_type_id* - referenced type ID, it might not exist yet;
2002
+ * Returns:
2003
+ * - >0, type ID of newly added BTF type;
2004
+ * - <0, on error.
2005
+ */
2006
+int btf__add_typedef(struct btf *btf, const char *name, int ref_type_id)
2007
+{
2008
+ if (!name || !name[0])
2009
+ return -EINVAL;
2010
+
2011
+ return btf_add_ref_kind(btf, BTF_KIND_TYPEDEF, name, ref_type_id);
2012
+}
2013
+
2014
+/*
2015
+ * Append new BTF_KIND_VOLATILE type with:
2016
+ * - *ref_type_id* - referenced type ID, it might not exist yet;
2017
+ * Returns:
2018
+ * - >0, type ID of newly added BTF type;
2019
+ * - <0, on error.
2020
+ */
2021
+int btf__add_volatile(struct btf *btf, int ref_type_id)
2022
+{
2023
+ return btf_add_ref_kind(btf, BTF_KIND_VOLATILE, NULL, ref_type_id);
2024
+}
2025
+
2026
+/*
2027
+ * Append new BTF_KIND_CONST type with:
2028
+ * - *ref_type_id* - referenced type ID, it might not exist yet;
2029
+ * Returns:
2030
+ * - >0, type ID of newly added BTF type;
2031
+ * - <0, on error.
2032
+ */
2033
+int btf__add_const(struct btf *btf, int ref_type_id)
2034
+{
2035
+ return btf_add_ref_kind(btf, BTF_KIND_CONST, NULL, ref_type_id);
2036
+}
2037
+
2038
+/*
2039
+ * Append new BTF_KIND_RESTRICT type with:
2040
+ * - *ref_type_id* - referenced type ID, it might not exist yet;
2041
+ * Returns:
2042
+ * - >0, type ID of newly added BTF type;
2043
+ * - <0, on error.
2044
+ */
2045
+int btf__add_restrict(struct btf *btf, int ref_type_id)
2046
+{
2047
+ return btf_add_ref_kind(btf, BTF_KIND_RESTRICT, NULL, ref_type_id);
2048
+}
2049
+
2050
+/*
2051
+ * Append new BTF_KIND_FUNC type with:
2052
+ * - *name*, non-empty/non-NULL name;
2053
+ * - *proto_type_id* - FUNC_PROTO's type ID, it might not exist yet;
2054
+ * Returns:
2055
+ * - >0, type ID of newly added BTF type;
2056
+ * - <0, on error.
2057
+ */
2058
+int btf__add_func(struct btf *btf, const char *name,
2059
+ enum btf_func_linkage linkage, int proto_type_id)
2060
+{
2061
+ int id;
2062
+
2063
+ if (!name || !name[0])
2064
+ return -EINVAL;
2065
+ if (linkage != BTF_FUNC_STATIC && linkage != BTF_FUNC_GLOBAL &&
2066
+ linkage != BTF_FUNC_EXTERN)
2067
+ return -EINVAL;
2068
+
2069
+ id = btf_add_ref_kind(btf, BTF_KIND_FUNC, name, proto_type_id);
2070
+ if (id > 0) {
2071
+ struct btf_type *t = btf_type_by_id(btf, id);
2072
+
2073
+ t->info = btf_type_info(BTF_KIND_FUNC, linkage, 0);
2074
+ }
2075
+ return id;
2076
+}
2077
+
2078
+/*
2079
+ * Append new BTF_KIND_FUNC_PROTO with:
2080
+ * - *ret_type_id* - type ID for return result of a function.
2081
+ *
2082
+ * Function prototype initially has no arguments, but they can be added by
2083
+ * btf__add_func_param() one by one, immediately after
2084
+ * btf__add_func_proto() succeeded.
2085
+ *
2086
+ * Returns:
2087
+ * - >0, type ID of newly added BTF type;
2088
+ * - <0, on error.
2089
+ */
2090
+int btf__add_func_proto(struct btf *btf, int ret_type_id)
2091
+{
2092
+ struct btf_type *t;
2093
+ int sz, err;
2094
+
2095
+ if (validate_type_id(ret_type_id))
2096
+ return -EINVAL;
2097
+
2098
+ if (btf_ensure_modifiable(btf))
2099
+ return -ENOMEM;
2100
+
2101
+ sz = sizeof(struct btf_type);
2102
+ t = btf_add_type_mem(btf, sz);
2103
+ if (!t)
2104
+ return -ENOMEM;
2105
+
2106
+ /* start out with vlen=0; this will be adjusted when adding enum
2107
+ * values, if necessary
2108
+ */
2109
+ t->name_off = 0;
2110
+ t->info = btf_type_info(BTF_KIND_FUNC_PROTO, 0, 0);
2111
+ t->type = ret_type_id;
2112
+
2113
+ err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
2114
+ if (err)
2115
+ return err;
2116
+
2117
+ btf->hdr->type_len += sz;
2118
+ btf->hdr->str_off += sz;
2119
+ btf->nr_types++;
2120
+ return btf->nr_types;
2121
+}
2122
+
2123
+/*
2124
+ * Append new function parameter for current FUNC_PROTO type with:
2125
+ * - *name* - parameter name, can be NULL or empty;
2126
+ * - *type_id* - type ID describing the type of the parameter.
2127
+ * Returns:
2128
+ * - 0, on success;
2129
+ * - <0, on error.
2130
+ */
2131
+int btf__add_func_param(struct btf *btf, const char *name, int type_id)
2132
+{
2133
+ struct btf_type *t;
2134
+ struct btf_param *p;
2135
+ int sz, name_off = 0;
2136
+
2137
+ if (validate_type_id(type_id))
2138
+ return -EINVAL;
2139
+
2140
+ /* last type should be BTF_KIND_FUNC_PROTO */
2141
+ if (btf->nr_types == 0)
2142
+ return -EINVAL;
2143
+ t = btf_type_by_id(btf, btf->nr_types);
2144
+ if (!btf_is_func_proto(t))
2145
+ return -EINVAL;
2146
+
2147
+ /* decompose and invalidate raw data */
2148
+ if (btf_ensure_modifiable(btf))
2149
+ return -ENOMEM;
2150
+
2151
+ sz = sizeof(struct btf_param);
2152
+ p = btf_add_type_mem(btf, sz);
2153
+ if (!p)
2154
+ return -ENOMEM;
2155
+
2156
+ if (name && name[0]) {
2157
+ name_off = btf__add_str(btf, name);
2158
+ if (name_off < 0)
2159
+ return name_off;
2160
+ }
2161
+
2162
+ p->name_off = name_off;
2163
+ p->type = type_id;
2164
+
2165
+ /* update parent type's vlen */
2166
+ t = btf_type_by_id(btf, btf->nr_types);
2167
+ btf_type_inc_vlen(t);
2168
+
2169
+ btf->hdr->type_len += sz;
2170
+ btf->hdr->str_off += sz;
2171
+ return 0;
2172
+}
2173
+
2174
+/*
2175
+ * Append new BTF_KIND_VAR type with:
2176
+ * - *name* - non-empty/non-NULL name;
2177
+ * - *linkage* - variable linkage, one of BTF_VAR_STATIC,
2178
+ * BTF_VAR_GLOBAL_ALLOCATED, or BTF_VAR_GLOBAL_EXTERN;
2179
+ * - *type_id* - type ID of the type describing the type of the variable.
2180
+ * Returns:
2181
+ * - >0, type ID of newly added BTF type;
2182
+ * - <0, on error.
2183
+ */
2184
+int btf__add_var(struct btf *btf, const char *name, int linkage, int type_id)
2185
+{
2186
+ struct btf_type *t;
2187
+ struct btf_var *v;
2188
+ int sz, err, name_off;
2189
+
2190
+ /* non-empty name */
2191
+ if (!name || !name[0])
2192
+ return -EINVAL;
2193
+ if (linkage != BTF_VAR_STATIC && linkage != BTF_VAR_GLOBAL_ALLOCATED &&
2194
+ linkage != BTF_VAR_GLOBAL_EXTERN)
2195
+ return -EINVAL;
2196
+ if (validate_type_id(type_id))
2197
+ return -EINVAL;
2198
+
2199
+ /* deconstruct BTF, if necessary, and invalidate raw_data */
2200
+ if (btf_ensure_modifiable(btf))
2201
+ return -ENOMEM;
2202
+
2203
+ sz = sizeof(struct btf_type) + sizeof(struct btf_var);
2204
+ t = btf_add_type_mem(btf, sz);
2205
+ if (!t)
2206
+ return -ENOMEM;
2207
+
2208
+ name_off = btf__add_str(btf, name);
2209
+ if (name_off < 0)
2210
+ return name_off;
2211
+
2212
+ t->name_off = name_off;
2213
+ t->info = btf_type_info(BTF_KIND_VAR, 0, 0);
2214
+ t->type = type_id;
2215
+
2216
+ v = btf_var(t);
2217
+ v->linkage = linkage;
2218
+
2219
+ err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
2220
+ if (err)
2221
+ return err;
2222
+
2223
+ btf->hdr->type_len += sz;
2224
+ btf->hdr->str_off += sz;
2225
+ btf->nr_types++;
2226
+ return btf->nr_types;
2227
+}
2228
+
2229
+/*
2230
+ * Append new BTF_KIND_DATASEC type with:
2231
+ * - *name* - non-empty/non-NULL name;
2232
+ * - *byte_sz* - data section size, in bytes.
2233
+ *
2234
+ * Data section is initially empty. Variables info can be added with
2235
+ * btf__add_datasec_var_info() calls, after btf__add_datasec() succeeds.
2236
+ *
2237
+ * Returns:
2238
+ * - >0, type ID of newly added BTF type;
2239
+ * - <0, on error.
2240
+ */
2241
+int btf__add_datasec(struct btf *btf, const char *name, __u32 byte_sz)
2242
+{
2243
+ struct btf_type *t;
2244
+ int sz, err, name_off;
2245
+
2246
+ /* non-empty name */
2247
+ if (!name || !name[0])
2248
+ return -EINVAL;
2249
+
2250
+ if (btf_ensure_modifiable(btf))
2251
+ return -ENOMEM;
2252
+
2253
+ sz = sizeof(struct btf_type);
2254
+ t = btf_add_type_mem(btf, sz);
2255
+ if (!t)
2256
+ return -ENOMEM;
2257
+
2258
+ name_off = btf__add_str(btf, name);
2259
+ if (name_off < 0)
2260
+ return name_off;
2261
+
2262
+ /* start with vlen=0, which will be update as var_secinfos are added */
2263
+ t->name_off = name_off;
2264
+ t->info = btf_type_info(BTF_KIND_DATASEC, 0, 0);
2265
+ t->size = byte_sz;
2266
+
2267
+ err = btf_add_type_idx_entry(btf, btf->hdr->type_len);
2268
+ if (err)
2269
+ return err;
2270
+
2271
+ btf->hdr->type_len += sz;
2272
+ btf->hdr->str_off += sz;
2273
+ btf->nr_types++;
2274
+ return btf->nr_types;
2275
+}
2276
+
2277
+/*
2278
+ * Append new data section variable information entry for current DATASEC type:
2279
+ * - *var_type_id* - type ID, describing type of the variable;
2280
+ * - *offset* - variable offset within data section, in bytes;
2281
+ * - *byte_sz* - variable size, in bytes.
2282
+ *
2283
+ * Returns:
2284
+ * - 0, on success;
2285
+ * - <0, on error.
2286
+ */
2287
+int btf__add_datasec_var_info(struct btf *btf, int var_type_id, __u32 offset, __u32 byte_sz)
2288
+{
2289
+ struct btf_type *t;
2290
+ struct btf_var_secinfo *v;
2291
+ int sz;
2292
+
2293
+ /* last type should be BTF_KIND_DATASEC */
2294
+ if (btf->nr_types == 0)
2295
+ return -EINVAL;
2296
+ t = btf_type_by_id(btf, btf->nr_types);
2297
+ if (!btf_is_datasec(t))
2298
+ return -EINVAL;
2299
+
2300
+ if (validate_type_id(var_type_id))
2301
+ return -EINVAL;
2302
+
2303
+ /* decompose and invalidate raw data */
2304
+ if (btf_ensure_modifiable(btf))
2305
+ return -ENOMEM;
2306
+
2307
+ sz = sizeof(struct btf_var_secinfo);
2308
+ v = btf_add_type_mem(btf, sz);
2309
+ if (!v)
2310
+ return -ENOMEM;
2311
+
2312
+ v->type = var_type_id;
2313
+ v->offset = offset;
2314
+ v->size = byte_sz;
2315
+
2316
+ /* update parent type's vlen */
2317
+ t = btf_type_by_id(btf, btf->nr_types);
2318
+ btf_type_inc_vlen(t);
2319
+
2320
+ btf->hdr->type_len += sz;
2321
+ btf->hdr->str_off += sz;
2322
+ return 0;
2323
+}
2324
+
2325
+struct btf_ext_sec_setup_param {
2326
+ __u32 off;
2327
+ __u32 len;
2328
+ __u32 min_rec_size;
2329
+ struct btf_ext_info *ext_info;
2330
+ const char *desc;
2331
+};
2332
+
2333
+static int btf_ext_setup_info(struct btf_ext *btf_ext,
2334
+ struct btf_ext_sec_setup_param *ext_sec)
2335
+{
2336
+ const struct btf_ext_info_sec *sinfo;
2337
+ struct btf_ext_info *ext_info;
2338
+ __u32 info_left, record_size;
2339
+ /* The start of the info sec (including the __u32 record_size). */
2340
+ void *info;
2341
+
2342
+ if (ext_sec->len == 0)
2343
+ return 0;
2344
+
2345
+ if (ext_sec->off & 0x03) {
2346
+ pr_debug(".BTF.ext %s section is not aligned to 4 bytes\n",
2347
+ ext_sec->desc);
2348
+ return -EINVAL;
2349
+ }
2350
+
2351
+ info = btf_ext->data + btf_ext->hdr->hdr_len + ext_sec->off;
2352
+ info_left = ext_sec->len;
2353
+
2354
+ if (btf_ext->data + btf_ext->data_size < info + ext_sec->len) {
2355
+ pr_debug("%s section (off:%u len:%u) is beyond the end of the ELF section .BTF.ext\n",
2356
+ ext_sec->desc, ext_sec->off, ext_sec->len);
2357
+ return -EINVAL;
2358
+ }
2359
+
2360
+ /* At least a record size */
2361
+ if (info_left < sizeof(__u32)) {
2362
+ pr_debug(".BTF.ext %s record size not found\n", ext_sec->desc);
2363
+ return -EINVAL;
2364
+ }
2365
+
2366
+ /* The record size needs to meet the minimum standard */
2367
+ record_size = *(__u32 *)info;
2368
+ if (record_size < ext_sec->min_rec_size ||
2369
+ record_size & 0x03) {
2370
+ pr_debug("%s section in .BTF.ext has invalid record size %u\n",
2371
+ ext_sec->desc, record_size);
2372
+ return -EINVAL;
2373
+ }
2374
+
2375
+ sinfo = info + sizeof(__u32);
2376
+ info_left -= sizeof(__u32);
2377
+
2378
+ /* If no records, return failure now so .BTF.ext won't be used. */
2379
+ if (!info_left) {
2380
+ pr_debug("%s section in .BTF.ext has no records", ext_sec->desc);
2381
+ return -EINVAL;
2382
+ }
2383
+
2384
+ while (info_left) {
2385
+ unsigned int sec_hdrlen = sizeof(struct btf_ext_info_sec);
2386
+ __u64 total_record_size;
2387
+ __u32 num_records;
2388
+
2389
+ if (info_left < sec_hdrlen) {
2390
+ pr_debug("%s section header is not found in .BTF.ext\n",
2391
+ ext_sec->desc);
2392
+ return -EINVAL;
2393
+ }
2394
+
2395
+ num_records = sinfo->num_info;
2396
+ if (num_records == 0) {
2397
+ pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2398
+ ext_sec->desc);
2399
+ return -EINVAL;
2400
+ }
2401
+
2402
+ total_record_size = sec_hdrlen +
2403
+ (__u64)num_records * record_size;
2404
+ if (info_left < total_record_size) {
2405
+ pr_debug("%s section has incorrect num_records in .BTF.ext\n",
2406
+ ext_sec->desc);
2407
+ return -EINVAL;
2408
+ }
2409
+
2410
+ info_left -= total_record_size;
2411
+ sinfo = (void *)sinfo + total_record_size;
2412
+ }
2413
+
2414
+ ext_info = ext_sec->ext_info;
2415
+ ext_info->len = ext_sec->len - sizeof(__u32);
2416
+ ext_info->rec_size = record_size;
2417
+ ext_info->info = info + sizeof(__u32);
2418
+
2419
+ return 0;
2420
+}
2421
+
2422
+static int btf_ext_setup_func_info(struct btf_ext *btf_ext)
2423
+{
2424
+ struct btf_ext_sec_setup_param param = {
2425
+ .off = btf_ext->hdr->func_info_off,
2426
+ .len = btf_ext->hdr->func_info_len,
2427
+ .min_rec_size = sizeof(struct bpf_func_info_min),
2428
+ .ext_info = &btf_ext->func_info,
2429
+ .desc = "func_info"
2430
+ };
2431
+
2432
+ return btf_ext_setup_info(btf_ext, &param);
2433
+}
2434
+
2435
+static int btf_ext_setup_line_info(struct btf_ext *btf_ext)
2436
+{
2437
+ struct btf_ext_sec_setup_param param = {
2438
+ .off = btf_ext->hdr->line_info_off,
2439
+ .len = btf_ext->hdr->line_info_len,
2440
+ .min_rec_size = sizeof(struct bpf_line_info_min),
2441
+ .ext_info = &btf_ext->line_info,
2442
+ .desc = "line_info",
2443
+ };
2444
+
2445
+ return btf_ext_setup_info(btf_ext, &param);
2446
+}
2447
+
2448
+static int btf_ext_setup_core_relos(struct btf_ext *btf_ext)
2449
+{
2450
+ struct btf_ext_sec_setup_param param = {
2451
+ .off = btf_ext->hdr->core_relo_off,
2452
+ .len = btf_ext->hdr->core_relo_len,
2453
+ .min_rec_size = sizeof(struct bpf_core_relo),
2454
+ .ext_info = &btf_ext->core_relo_info,
2455
+ .desc = "core_relo",
2456
+ };
2457
+
2458
+ return btf_ext_setup_info(btf_ext, &param);
2459
+}
2460
+
2461
+static int btf_ext_parse_hdr(__u8 *data, __u32 data_size)
2462
+{
2463
+ const struct btf_ext_header *hdr = (struct btf_ext_header *)data;
2464
+
2465
+ if (data_size < offsetofend(struct btf_ext_header, hdr_len) ||
2466
+ data_size < hdr->hdr_len) {
2467
+ pr_debug("BTF.ext header not found");
2468
+ return -EINVAL;
2469
+ }
2470
+
2471
+ if (hdr->magic == bswap_16(BTF_MAGIC)) {
2472
+ pr_warn("BTF.ext in non-native endianness is not supported\n");
2473
+ return -ENOTSUP;
2474
+ } else if (hdr->magic != BTF_MAGIC) {
2475
+ pr_debug("Invalid BTF.ext magic:%x\n", hdr->magic);
2476
+ return -EINVAL;
2477
+ }
2478
+
2479
+ if (hdr->version != BTF_VERSION) {
2480
+ pr_debug("Unsupported BTF.ext version:%u\n", hdr->version);
2481
+ return -ENOTSUP;
2482
+ }
2483
+
2484
+ if (hdr->flags) {
2485
+ pr_debug("Unsupported BTF.ext flags:%x\n", hdr->flags);
2486
+ return -ENOTSUP;
2487
+ }
2488
+
2489
+ if (data_size == hdr->hdr_len) {
2490
+ pr_debug("BTF.ext has no data\n");
2491
+ return -EINVAL;
2492
+ }
2493
+
2494
+ return 0;
2495
+}
2496
+
2497
+void btf_ext__free(struct btf_ext *btf_ext)
2498
+{
2499
+ if (IS_ERR_OR_NULL(btf_ext))
2500
+ return;
2501
+ free(btf_ext->data);
2502
+ free(btf_ext);
2503
+}
2504
+
2505
+struct btf_ext *btf_ext__new(__u8 *data, __u32 size)
2506
+{
2507
+ struct btf_ext *btf_ext;
2508
+ int err;
2509
+
2510
+ err = btf_ext_parse_hdr(data, size);
2511
+ if (err)
2512
+ return ERR_PTR(err);
2513
+
2514
+ btf_ext = calloc(1, sizeof(struct btf_ext));
2515
+ if (!btf_ext)
2516
+ return ERR_PTR(-ENOMEM);
2517
+
2518
+ btf_ext->data_size = size;
2519
+ btf_ext->data = malloc(size);
2520
+ if (!btf_ext->data) {
2521
+ err = -ENOMEM;
2522
+ goto done;
2523
+ }
2524
+ memcpy(btf_ext->data, data, size);
2525
+
2526
+ if (btf_ext->hdr->hdr_len <
2527
+ offsetofend(struct btf_ext_header, line_info_len))
2528
+ goto done;
2529
+ err = btf_ext_setup_func_info(btf_ext);
2530
+ if (err)
2531
+ goto done;
2532
+
2533
+ err = btf_ext_setup_line_info(btf_ext);
2534
+ if (err)
2535
+ goto done;
2536
+
2537
+ if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, core_relo_len))
2538
+ goto done;
2539
+ err = btf_ext_setup_core_relos(btf_ext);
2540
+ if (err)
2541
+ goto done;
2542
+
2543
+done:
2544
+ if (err) {
2545
+ btf_ext__free(btf_ext);
2546
+ return ERR_PTR(err);
2547
+ }
2548
+
2549
+ return btf_ext;
2550
+}
2551
+
2552
+const void *btf_ext__get_raw_data(const struct btf_ext *btf_ext, __u32 *size)
2553
+{
2554
+ *size = btf_ext->data_size;
2555
+ return btf_ext->data;
2556
+}
2557
+
2558
+static int btf_ext_reloc_info(const struct btf *btf,
2559
+ const struct btf_ext_info *ext_info,
2560
+ const char *sec_name, __u32 insns_cnt,
2561
+ void **info, __u32 *cnt)
2562
+{
2563
+ __u32 sec_hdrlen = sizeof(struct btf_ext_info_sec);
2564
+ __u32 i, record_size, existing_len, records_len;
2565
+ struct btf_ext_info_sec *sinfo;
2566
+ const char *info_sec_name;
2567
+ __u64 remain_len;
2568
+ void *data;
2569
+
2570
+ record_size = ext_info->rec_size;
2571
+ sinfo = ext_info->info;
2572
+ remain_len = ext_info->len;
2573
+ while (remain_len > 0) {
2574
+ records_len = sinfo->num_info * record_size;
2575
+ info_sec_name = btf__name_by_offset(btf, sinfo->sec_name_off);
2576
+ if (strcmp(info_sec_name, sec_name)) {
2577
+ remain_len -= sec_hdrlen + records_len;
2578
+ sinfo = (void *)sinfo + sec_hdrlen + records_len;
2579
+ continue;
2580
+ }
2581
+
2582
+ existing_len = (*cnt) * record_size;
2583
+ data = realloc(*info, existing_len + records_len);
2584
+ if (!data)
2585
+ return -ENOMEM;
2586
+
2587
+ memcpy(data + existing_len, sinfo->data, records_len);
2588
+ /* adjust insn_off only, the rest data will be passed
2589
+ * to the kernel.
2590
+ */
2591
+ for (i = 0; i < sinfo->num_info; i++) {
2592
+ __u32 *insn_off;
2593
+
2594
+ insn_off = data + existing_len + (i * record_size);
2595
+ *insn_off = *insn_off / sizeof(struct bpf_insn) +
2596
+ insns_cnt;
2597
+ }
2598
+ *info = data;
2599
+ *cnt += sinfo->num_info;
2600
+ return 0;
2601
+ }
2602
+
2603
+ return -ENOENT;
2604
+}
2605
+
2606
+int btf_ext__reloc_func_info(const struct btf *btf,
2607
+ const struct btf_ext *btf_ext,
2608
+ const char *sec_name, __u32 insns_cnt,
2609
+ void **func_info, __u32 *cnt)
2610
+{
2611
+ return btf_ext_reloc_info(btf, &btf_ext->func_info, sec_name,
2612
+ insns_cnt, func_info, cnt);
2613
+}
2614
+
2615
+int btf_ext__reloc_line_info(const struct btf *btf,
2616
+ const struct btf_ext *btf_ext,
2617
+ const char *sec_name, __u32 insns_cnt,
2618
+ void **line_info, __u32 *cnt)
2619
+{
2620
+ return btf_ext_reloc_info(btf, &btf_ext->line_info, sec_name,
2621
+ insns_cnt, line_info, cnt);
2622
+}
2623
+
2624
+__u32 btf_ext__func_info_rec_size(const struct btf_ext *btf_ext)
2625
+{
2626
+ return btf_ext->func_info.rec_size;
2627
+}
2628
+
2629
+__u32 btf_ext__line_info_rec_size(const struct btf_ext *btf_ext)
2630
+{
2631
+ return btf_ext->line_info.rec_size;
2632
+}
2633
+
2634
+struct btf_dedup;
2635
+
2636
+static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
2637
+ const struct btf_dedup_opts *opts);
2638
+static void btf_dedup_free(struct btf_dedup *d);
2639
+static int btf_dedup_strings(struct btf_dedup *d);
2640
+static int btf_dedup_prim_types(struct btf_dedup *d);
2641
+static int btf_dedup_struct_types(struct btf_dedup *d);
2642
+static int btf_dedup_ref_types(struct btf_dedup *d);
2643
+static int btf_dedup_compact_types(struct btf_dedup *d);
2644
+static int btf_dedup_remap_types(struct btf_dedup *d);
2645
+
2646
+/*
2647
+ * Deduplicate BTF types and strings.
2648
+ *
2649
+ * BTF dedup algorithm takes as an input `struct btf` representing `.BTF` ELF
2650
+ * section with all BTF type descriptors and string data. It overwrites that
2651
+ * memory in-place with deduplicated types and strings without any loss of
2652
+ * information. If optional `struct btf_ext` representing '.BTF.ext' ELF section
2653
+ * is provided, all the strings referenced from .BTF.ext section are honored
2654
+ * and updated to point to the right offsets after deduplication.
2655
+ *
2656
+ * If function returns with error, type/string data might be garbled and should
2657
+ * be discarded.
2658
+ *
2659
+ * More verbose and detailed description of both problem btf_dedup is solving,
2660
+ * as well as solution could be found at:
2661
+ * https://facebookmicrosites.github.io/bpf/blog/2018/11/14/btf-enhancement.html
2662
+ *
2663
+ * Problem description and justification
2664
+ * =====================================
2665
+ *
2666
+ * BTF type information is typically emitted either as a result of conversion
2667
+ * from DWARF to BTF or directly by compiler. In both cases, each compilation
2668
+ * unit contains information about a subset of all the types that are used
2669
+ * in an application. These subsets are frequently overlapping and contain a lot
2670
+ * of duplicated information when later concatenated together into a single
2671
+ * binary. This algorithm ensures that each unique type is represented by single
2672
+ * BTF type descriptor, greatly reducing resulting size of BTF data.
2673
+ *
2674
+ * Compilation unit isolation and subsequent duplication of data is not the only
2675
+ * problem. The same type hierarchy (e.g., struct and all the type that struct
2676
+ * references) in different compilation units can be represented in BTF to
2677
+ * various degrees of completeness (or, rather, incompleteness) due to
2678
+ * struct/union forward declarations.
2679
+ *
2680
+ * Let's take a look at an example, that we'll use to better understand the
2681
+ * problem (and solution). Suppose we have two compilation units, each using
2682
+ * same `struct S`, but each of them having incomplete type information about
2683
+ * struct's fields:
2684
+ *
2685
+ * // CU #1:
2686
+ * struct S;
2687
+ * struct A {
2688
+ * int a;
2689
+ * struct A* self;
2690
+ * struct S* parent;
2691
+ * };
2692
+ * struct B;
2693
+ * struct S {
2694
+ * struct A* a_ptr;
2695
+ * struct B* b_ptr;
2696
+ * };
2697
+ *
2698
+ * // CU #2:
2699
+ * struct S;
2700
+ * struct A;
2701
+ * struct B {
2702
+ * int b;
2703
+ * struct B* self;
2704
+ * struct S* parent;
2705
+ * };
2706
+ * struct S {
2707
+ * struct A* a_ptr;
2708
+ * struct B* b_ptr;
2709
+ * };
2710
+ *
2711
+ * In case of CU #1, BTF data will know only that `struct B` exist (but no
2712
+ * more), but will know the complete type information about `struct A`. While
2713
+ * for CU #2, it will know full type information about `struct B`, but will
2714
+ * only know about forward declaration of `struct A` (in BTF terms, it will
2715
+ * have `BTF_KIND_FWD` type descriptor with name `B`).
2716
+ *
2717
+ * This compilation unit isolation means that it's possible that there is no
2718
+ * single CU with complete type information describing structs `S`, `A`, and
2719
+ * `B`. Also, we might get tons of duplicated and redundant type information.
2720
+ *
2721
+ * Additional complication we need to keep in mind comes from the fact that
2722
+ * types, in general, can form graphs containing cycles, not just DAGs.
2723
+ *
2724
+ * While algorithm does deduplication, it also merges and resolves type
2725
+ * information (unless disabled throught `struct btf_opts`), whenever possible.
2726
+ * E.g., in the example above with two compilation units having partial type
2727
+ * information for structs `A` and `B`, the output of algorithm will emit
2728
+ * a single copy of each BTF type that describes structs `A`, `B`, and `S`
2729
+ * (as well as type information for `int` and pointers), as if they were defined
2730
+ * in a single compilation unit as:
2731
+ *
2732
+ * struct A {
2733
+ * int a;
2734
+ * struct A* self;
2735
+ * struct S* parent;
2736
+ * };
2737
+ * struct B {
2738
+ * int b;
2739
+ * struct B* self;
2740
+ * struct S* parent;
2741
+ * };
2742
+ * struct S {
2743
+ * struct A* a_ptr;
2744
+ * struct B* b_ptr;
2745
+ * };
2746
+ *
2747
+ * Algorithm summary
2748
+ * =================
2749
+ *
2750
+ * Algorithm completes its work in 6 separate passes:
2751
+ *
2752
+ * 1. Strings deduplication.
2753
+ * 2. Primitive types deduplication (int, enum, fwd).
2754
+ * 3. Struct/union types deduplication.
2755
+ * 4. Reference types deduplication (pointers, typedefs, arrays, funcs, func
2756
+ * protos, and const/volatile/restrict modifiers).
2757
+ * 5. Types compaction.
2758
+ * 6. Types remapping.
2759
+ *
2760
+ * Algorithm determines canonical type descriptor, which is a single
2761
+ * representative type for each truly unique type. This canonical type is the
2762
+ * one that will go into final deduplicated BTF type information. For
2763
+ * struct/unions, it is also the type that algorithm will merge additional type
2764
+ * information into (while resolving FWDs), as it discovers it from data in
2765
+ * other CUs. Each input BTF type eventually gets either mapped to itself, if
2766
+ * that type is canonical, or to some other type, if that type is equivalent
2767
+ * and was chosen as canonical representative. This mapping is stored in
2768
+ * `btf_dedup->map` array. This map is also used to record STRUCT/UNION that
2769
+ * FWD type got resolved to.
2770
+ *
2771
+ * To facilitate fast discovery of canonical types, we also maintain canonical
2772
+ * index (`btf_dedup->dedup_table`), which maps type descriptor's signature hash
2773
+ * (i.e., hashed kind, name, size, fields, etc) into a list of canonical types
2774
+ * that match that signature. With sufficiently good choice of type signature
2775
+ * hashing function, we can limit number of canonical types for each unique type
2776
+ * signature to a very small number, allowing to find canonical type for any
2777
+ * duplicated type very quickly.
2778
+ *
2779
+ * Struct/union deduplication is the most critical part and algorithm for
2780
+ * deduplicating structs/unions is described in greater details in comments for
2781
+ * `btf_dedup_is_equiv` function.
2782
+ */
2783
+int btf__dedup(struct btf *btf, struct btf_ext *btf_ext,
2784
+ const struct btf_dedup_opts *opts)
2785
+{
2786
+ struct btf_dedup *d = btf_dedup_new(btf, btf_ext, opts);
2787
+ int err;
2788
+
2789
+ if (IS_ERR(d)) {
2790
+ pr_debug("btf_dedup_new failed: %ld", PTR_ERR(d));
2791
+ return -EINVAL;
2792
+ }
2793
+
2794
+ if (btf_ensure_modifiable(btf))
2795
+ return -ENOMEM;
2796
+
2797
+ err = btf_dedup_strings(d);
2798
+ if (err < 0) {
2799
+ pr_debug("btf_dedup_strings failed:%d\n", err);
2800
+ goto done;
2801
+ }
2802
+ err = btf_dedup_prim_types(d);
2803
+ if (err < 0) {
2804
+ pr_debug("btf_dedup_prim_types failed:%d\n", err);
2805
+ goto done;
2806
+ }
2807
+ err = btf_dedup_struct_types(d);
2808
+ if (err < 0) {
2809
+ pr_debug("btf_dedup_struct_types failed:%d\n", err);
2810
+ goto done;
2811
+ }
2812
+ err = btf_dedup_ref_types(d);
2813
+ if (err < 0) {
2814
+ pr_debug("btf_dedup_ref_types failed:%d\n", err);
2815
+ goto done;
2816
+ }
2817
+ err = btf_dedup_compact_types(d);
2818
+ if (err < 0) {
2819
+ pr_debug("btf_dedup_compact_types failed:%d\n", err);
2820
+ goto done;
2821
+ }
2822
+ err = btf_dedup_remap_types(d);
2823
+ if (err < 0) {
2824
+ pr_debug("btf_dedup_remap_types failed:%d\n", err);
2825
+ goto done;
2826
+ }
2827
+
2828
+done:
2829
+ btf_dedup_free(d);
2830
+ return err;
2831
+}
2832
+
2833
+#define BTF_UNPROCESSED_ID ((__u32)-1)
2834
+#define BTF_IN_PROGRESS_ID ((__u32)-2)
2835
+
2836
+struct btf_dedup {
2837
+ /* .BTF section to be deduped in-place */
2838
+ struct btf *btf;
2839
+ /*
2840
+ * Optional .BTF.ext section. When provided, any strings referenced
2841
+ * from it will be taken into account when deduping strings
2842
+ */
2843
+ struct btf_ext *btf_ext;
2844
+ /*
2845
+ * This is a map from any type's signature hash to a list of possible
2846
+ * canonical representative type candidates. Hash collisions are
2847
+ * ignored, so even types of various kinds can share same list of
2848
+ * candidates, which is fine because we rely on subsequent
2849
+ * btf_xxx_equal() checks to authoritatively verify type equality.
2850
+ */
2851
+ struct hashmap *dedup_table;
2852
+ /* Canonical types map */
2853
+ __u32 *map;
2854
+ /* Hypothetical mapping, used during type graph equivalence checks */
2855
+ __u32 *hypot_map;
2856
+ __u32 *hypot_list;
2857
+ size_t hypot_cnt;
2858
+ size_t hypot_cap;
2859
+ /* Various option modifying behavior of algorithm */
2860
+ struct btf_dedup_opts opts;
2861
+};
2862
+
2863
+struct btf_str_ptr {
2864
+ const char *str;
2865
+ __u32 new_off;
2866
+ bool used;
2867
+};
2868
+
2869
+struct btf_str_ptrs {
2870
+ struct btf_str_ptr *ptrs;
2871
+ const char *data;
2872
+ __u32 cnt;
2873
+ __u32 cap;
2874
+};
2875
+
2876
+static long hash_combine(long h, long value)
2877
+{
2878
+ return h * 31 + value;
2879
+}
2880
+
2881
+#define for_each_dedup_cand(d, node, hash) \
2882
+ hashmap__for_each_key_entry(d->dedup_table, node, (void *)hash)
2883
+
2884
+static int btf_dedup_table_add(struct btf_dedup *d, long hash, __u32 type_id)
2885
+{
2886
+ return hashmap__append(d->dedup_table,
2887
+ (void *)hash, (void *)(long)type_id);
2888
+}
2889
+
2890
+static int btf_dedup_hypot_map_add(struct btf_dedup *d,
2891
+ __u32 from_id, __u32 to_id)
2892
+{
2893
+ if (d->hypot_cnt == d->hypot_cap) {
2894
+ __u32 *new_list;
2895
+
2896
+ d->hypot_cap += max((size_t)16, d->hypot_cap / 2);
2897
+ new_list = libbpf_reallocarray(d->hypot_list, d->hypot_cap, sizeof(__u32));
2898
+ if (!new_list)
2899
+ return -ENOMEM;
2900
+ d->hypot_list = new_list;
2901
+ }
2902
+ d->hypot_list[d->hypot_cnt++] = from_id;
2903
+ d->hypot_map[from_id] = to_id;
2904
+ return 0;
2905
+}
2906
+
2907
+static void btf_dedup_clear_hypot_map(struct btf_dedup *d)
2908
+{
2909
+ int i;
2910
+
2911
+ for (i = 0; i < d->hypot_cnt; i++)
2912
+ d->hypot_map[d->hypot_list[i]] = BTF_UNPROCESSED_ID;
2913
+ d->hypot_cnt = 0;
2914
+}
2915
+
2916
+static void btf_dedup_free(struct btf_dedup *d)
2917
+{
2918
+ hashmap__free(d->dedup_table);
2919
+ d->dedup_table = NULL;
2920
+
2921
+ free(d->map);
2922
+ d->map = NULL;
2923
+
2924
+ free(d->hypot_map);
2925
+ d->hypot_map = NULL;
2926
+
2927
+ free(d->hypot_list);
2928
+ d->hypot_list = NULL;
2929
+
2930
+ free(d);
2931
+}
2932
+
2933
+static size_t btf_dedup_identity_hash_fn(const void *key, void *ctx)
2934
+{
2935
+ return (size_t)key;
2936
+}
2937
+
2938
+static size_t btf_dedup_collision_hash_fn(const void *key, void *ctx)
2939
+{
2940
+ return 0;
2941
+}
2942
+
2943
+static bool btf_dedup_equal_fn(const void *k1, const void *k2, void *ctx)
2944
+{
2945
+ return k1 == k2;
2946
+}
2947
+
2948
+static struct btf_dedup *btf_dedup_new(struct btf *btf, struct btf_ext *btf_ext,
2949
+ const struct btf_dedup_opts *opts)
2950
+{
2951
+ struct btf_dedup *d = calloc(1, sizeof(struct btf_dedup));
2952
+ hashmap_hash_fn hash_fn = btf_dedup_identity_hash_fn;
2953
+ int i, err = 0;
2954
+
2955
+ if (!d)
2956
+ return ERR_PTR(-ENOMEM);
2957
+
2958
+ d->opts.dont_resolve_fwds = opts && opts->dont_resolve_fwds;
2959
+ /* dedup_table_size is now used only to force collisions in tests */
2960
+ if (opts && opts->dedup_table_size == 1)
2961
+ hash_fn = btf_dedup_collision_hash_fn;
2962
+
2963
+ d->btf = btf;
2964
+ d->btf_ext = btf_ext;
2965
+
2966
+ d->dedup_table = hashmap__new(hash_fn, btf_dedup_equal_fn, NULL);
2967
+ if (IS_ERR(d->dedup_table)) {
2968
+ err = PTR_ERR(d->dedup_table);
2969
+ d->dedup_table = NULL;
2970
+ goto done;
2971
+ }
2972
+
2973
+ d->map = malloc(sizeof(__u32) * (1 + btf->nr_types));
2974
+ if (!d->map) {
2975
+ err = -ENOMEM;
2976
+ goto done;
2977
+ }
2978
+ /* special BTF "void" type is made canonical immediately */
2979
+ d->map[0] = 0;
2980
+ for (i = 1; i <= btf->nr_types; i++) {
2981
+ struct btf_type *t = btf_type_by_id(d->btf, i);
2982
+
2983
+ /* VAR and DATASEC are never deduped and are self-canonical */
2984
+ if (btf_is_var(t) || btf_is_datasec(t))
2985
+ d->map[i] = i;
2986
+ else
2987
+ d->map[i] = BTF_UNPROCESSED_ID;
2988
+ }
2989
+
2990
+ d->hypot_map = malloc(sizeof(__u32) * (1 + btf->nr_types));
2991
+ if (!d->hypot_map) {
2992
+ err = -ENOMEM;
2993
+ goto done;
2994
+ }
2995
+ for (i = 0; i <= btf->nr_types; i++)
2996
+ d->hypot_map[i] = BTF_UNPROCESSED_ID;
2997
+
2998
+done:
2999
+ if (err) {
3000
+ btf_dedup_free(d);
3001
+ return ERR_PTR(err);
3002
+ }
3003
+
3004
+ return d;
3005
+}
3006
+
3007
+typedef int (*str_off_fn_t)(__u32 *str_off_ptr, void *ctx);
3008
+
3009
+/*
3010
+ * Iterate over all possible places in .BTF and .BTF.ext that can reference
3011
+ * string and pass pointer to it to a provided callback `fn`.
3012
+ */
3013
+static int btf_for_each_str_off(struct btf_dedup *d, str_off_fn_t fn, void *ctx)
3014
+{
3015
+ void *line_data_cur, *line_data_end;
3016
+ int i, j, r, rec_size;
3017
+ struct btf_type *t;
3018
+
3019
+ for (i = 1; i <= d->btf->nr_types; i++) {
3020
+ t = btf_type_by_id(d->btf, i);
3021
+ r = fn(&t->name_off, ctx);
3022
+ if (r)
3023
+ return r;
3024
+
3025
+ switch (btf_kind(t)) {
3026
+ case BTF_KIND_STRUCT:
3027
+ case BTF_KIND_UNION: {
3028
+ struct btf_member *m = btf_members(t);
3029
+ __u16 vlen = btf_vlen(t);
3030
+
3031
+ for (j = 0; j < vlen; j++) {
3032
+ r = fn(&m->name_off, ctx);
3033
+ if (r)
3034
+ return r;
3035
+ m++;
3036
+ }
3037
+ break;
3038
+ }
3039
+ case BTF_KIND_ENUM: {
3040
+ struct btf_enum *m = btf_enum(t);
3041
+ __u16 vlen = btf_vlen(t);
3042
+
3043
+ for (j = 0; j < vlen; j++) {
3044
+ r = fn(&m->name_off, ctx);
3045
+ if (r)
3046
+ return r;
3047
+ m++;
3048
+ }
3049
+ break;
3050
+ }
3051
+ case BTF_KIND_FUNC_PROTO: {
3052
+ struct btf_param *m = btf_params(t);
3053
+ __u16 vlen = btf_vlen(t);
3054
+
3055
+ for (j = 0; j < vlen; j++) {
3056
+ r = fn(&m->name_off, ctx);
3057
+ if (r)
3058
+ return r;
3059
+ m++;
3060
+ }
3061
+ break;
3062
+ }
3063
+ default:
3064
+ break;
3065
+ }
3066
+ }
3067
+
3068
+ if (!d->btf_ext)
3069
+ return 0;
3070
+
3071
+ line_data_cur = d->btf_ext->line_info.info;
3072
+ line_data_end = d->btf_ext->line_info.info + d->btf_ext->line_info.len;
3073
+ rec_size = d->btf_ext->line_info.rec_size;
3074
+
3075
+ while (line_data_cur < line_data_end) {
3076
+ struct btf_ext_info_sec *sec = line_data_cur;
3077
+ struct bpf_line_info_min *line_info;
3078
+ __u32 num_info = sec->num_info;
3079
+
3080
+ r = fn(&sec->sec_name_off, ctx);
3081
+ if (r)
3082
+ return r;
3083
+
3084
+ line_data_cur += sizeof(struct btf_ext_info_sec);
3085
+ for (i = 0; i < num_info; i++) {
3086
+ line_info = line_data_cur;
3087
+ r = fn(&line_info->file_name_off, ctx);
3088
+ if (r)
3089
+ return r;
3090
+ r = fn(&line_info->line_off, ctx);
3091
+ if (r)
3092
+ return r;
3093
+ line_data_cur += rec_size;
3094
+ }
3095
+ }
3096
+
3097
+ return 0;
3098
+}
3099
+
3100
+static int str_sort_by_content(const void *a1, const void *a2)
3101
+{
3102
+ const struct btf_str_ptr *p1 = a1;
3103
+ const struct btf_str_ptr *p2 = a2;
3104
+
3105
+ return strcmp(p1->str, p2->str);
3106
+}
3107
+
3108
+static int str_sort_by_offset(const void *a1, const void *a2)
3109
+{
3110
+ const struct btf_str_ptr *p1 = a1;
3111
+ const struct btf_str_ptr *p2 = a2;
3112
+
3113
+ if (p1->str != p2->str)
3114
+ return p1->str < p2->str ? -1 : 1;
3115
+ return 0;
3116
+}
3117
+
3118
+static int btf_dedup_str_ptr_cmp(const void *str_ptr, const void *pelem)
3119
+{
3120
+ const struct btf_str_ptr *p = pelem;
3121
+
3122
+ if (str_ptr != p->str)
3123
+ return (const char *)str_ptr < p->str ? -1 : 1;
3124
+ return 0;
3125
+}
3126
+
3127
+static int btf_str_mark_as_used(__u32 *str_off_ptr, void *ctx)
3128
+{
3129
+ struct btf_str_ptrs *strs;
3130
+ struct btf_str_ptr *s;
3131
+
3132
+ if (*str_off_ptr == 0)
3133
+ return 0;
3134
+
3135
+ strs = ctx;
3136
+ s = bsearch(strs->data + *str_off_ptr, strs->ptrs, strs->cnt,
3137
+ sizeof(struct btf_str_ptr), btf_dedup_str_ptr_cmp);
3138
+ if (!s)
3139
+ return -EINVAL;
3140
+ s->used = true;
3141
+ return 0;
3142
+}
3143
+
3144
+static int btf_str_remap_offset(__u32 *str_off_ptr, void *ctx)
3145
+{
3146
+ struct btf_str_ptrs *strs;
3147
+ struct btf_str_ptr *s;
3148
+
3149
+ if (*str_off_ptr == 0)
3150
+ return 0;
3151
+
3152
+ strs = ctx;
3153
+ s = bsearch(strs->data + *str_off_ptr, strs->ptrs, strs->cnt,
3154
+ sizeof(struct btf_str_ptr), btf_dedup_str_ptr_cmp);
3155
+ if (!s)
3156
+ return -EINVAL;
3157
+ *str_off_ptr = s->new_off;
3158
+ return 0;
3159
+}
3160
+
3161
+/*
3162
+ * Dedup string and filter out those that are not referenced from either .BTF
3163
+ * or .BTF.ext (if provided) sections.
3164
+ *
3165
+ * This is done by building index of all strings in BTF's string section,
3166
+ * then iterating over all entities that can reference strings (e.g., type
3167
+ * names, struct field names, .BTF.ext line info, etc) and marking corresponding
3168
+ * strings as used. After that all used strings are deduped and compacted into
3169
+ * sequential blob of memory and new offsets are calculated. Then all the string
3170
+ * references are iterated again and rewritten using new offsets.
3171
+ */
3172
+static int btf_dedup_strings(struct btf_dedup *d)
3173
+{
3174
+ char *start = d->btf->strs_data;
3175
+ char *end = start + d->btf->hdr->str_len;
3176
+ char *p = start, *tmp_strs = NULL;
3177
+ struct btf_str_ptrs strs = {
3178
+ .cnt = 0,
3179
+ .cap = 0,
3180
+ .ptrs = NULL,
3181
+ .data = start,
3182
+ };
3183
+ int i, j, err = 0, grp_idx;
3184
+ bool grp_used;
3185
+
3186
+ if (d->btf->strs_deduped)
3187
+ return 0;
3188
+
3189
+ /* build index of all strings */
3190
+ while (p < end) {
3191
+ if (strs.cnt + 1 > strs.cap) {
3192
+ struct btf_str_ptr *new_ptrs;
3193
+
3194
+ strs.cap += max(strs.cnt / 2, 16U);
3195
+ new_ptrs = libbpf_reallocarray(strs.ptrs, strs.cap, sizeof(strs.ptrs[0]));
3196
+ if (!new_ptrs) {
3197
+ err = -ENOMEM;
3198
+ goto done;
3199
+ }
3200
+ strs.ptrs = new_ptrs;
3201
+ }
3202
+
3203
+ strs.ptrs[strs.cnt].str = p;
3204
+ strs.ptrs[strs.cnt].used = false;
3205
+
3206
+ p += strlen(p) + 1;
3207
+ strs.cnt++;
3208
+ }
3209
+
3210
+ /* temporary storage for deduplicated strings */
3211
+ tmp_strs = malloc(d->btf->hdr->str_len);
3212
+ if (!tmp_strs) {
3213
+ err = -ENOMEM;
3214
+ goto done;
3215
+ }
3216
+
3217
+ /* mark all used strings */
3218
+ strs.ptrs[0].used = true;
3219
+ err = btf_for_each_str_off(d, btf_str_mark_as_used, &strs);
3220
+ if (err)
3221
+ goto done;
3222
+
3223
+ /* sort strings by context, so that we can identify duplicates */
3224
+ qsort(strs.ptrs, strs.cnt, sizeof(strs.ptrs[0]), str_sort_by_content);
3225
+
3226
+ /*
3227
+ * iterate groups of equal strings and if any instance in a group was
3228
+ * referenced, emit single instance and remember new offset
3229
+ */
3230
+ p = tmp_strs;
3231
+ grp_idx = 0;
3232
+ grp_used = strs.ptrs[0].used;
3233
+ /* iterate past end to avoid code duplication after loop */
3234
+ for (i = 1; i <= strs.cnt; i++) {
3235
+ /*
3236
+ * when i == strs.cnt, we want to skip string comparison and go
3237
+ * straight to handling last group of strings (otherwise we'd
3238
+ * need to handle last group after the loop w/ duplicated code)
3239
+ */
3240
+ if (i < strs.cnt &&
3241
+ !strcmp(strs.ptrs[i].str, strs.ptrs[grp_idx].str)) {
3242
+ grp_used = grp_used || strs.ptrs[i].used;
3243
+ continue;
3244
+ }
3245
+
3246
+ /*
3247
+ * this check would have been required after the loop to handle
3248
+ * last group of strings, but due to <= condition in a loop
3249
+ * we avoid that duplication
3250
+ */
3251
+ if (grp_used) {
3252
+ int new_off = p - tmp_strs;
3253
+ __u32 len = strlen(strs.ptrs[grp_idx].str);
3254
+
3255
+ memmove(p, strs.ptrs[grp_idx].str, len + 1);
3256
+ for (j = grp_idx; j < i; j++)
3257
+ strs.ptrs[j].new_off = new_off;
3258
+ p += len + 1;
3259
+ }
3260
+
3261
+ if (i < strs.cnt) {
3262
+ grp_idx = i;
3263
+ grp_used = strs.ptrs[i].used;
3264
+ }
3265
+ }
3266
+
3267
+ /* replace original strings with deduped ones */
3268
+ d->btf->hdr->str_len = p - tmp_strs;
3269
+ memmove(start, tmp_strs, d->btf->hdr->str_len);
3270
+ end = start + d->btf->hdr->str_len;
3271
+
3272
+ /* restore original order for further binary search lookups */
3273
+ qsort(strs.ptrs, strs.cnt, sizeof(strs.ptrs[0]), str_sort_by_offset);
3274
+
3275
+ /* remap string offsets */
3276
+ err = btf_for_each_str_off(d, btf_str_remap_offset, &strs);
3277
+ if (err)
3278
+ goto done;
3279
+
3280
+ d->btf->hdr->str_len = end - start;
3281
+ d->btf->strs_deduped = true;
3282
+
3283
+done:
3284
+ free(tmp_strs);
3285
+ free(strs.ptrs);
3286
+ return err;
3287
+}
3288
+
3289
+static long btf_hash_common(struct btf_type *t)
3290
+{
3291
+ long h;
3292
+
3293
+ h = hash_combine(0, t->name_off);
3294
+ h = hash_combine(h, t->info);
3295
+ h = hash_combine(h, t->size);
3296
+ return h;
3297
+}
3298
+
3299
+static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
3300
+{
3301
+ return t1->name_off == t2->name_off &&
3302
+ t1->info == t2->info &&
3303
+ t1->size == t2->size;
3304
+}
3305
+
3306
+/* Calculate type signature hash of INT. */
3307
+static long btf_hash_int(struct btf_type *t)
3308
+{
3309
+ __u32 info = *(__u32 *)(t + 1);
3310
+ long h;
3311
+
3312
+ h = btf_hash_common(t);
3313
+ h = hash_combine(h, info);
3314
+ return h;
3315
+}
3316
+
3317
+/* Check structural equality of two INTs. */
3318
+static bool btf_equal_int(struct btf_type *t1, struct btf_type *t2)
3319
+{
3320
+ __u32 info1, info2;
3321
+
3322
+ if (!btf_equal_common(t1, t2))
3323
+ return false;
3324
+ info1 = *(__u32 *)(t1 + 1);
3325
+ info2 = *(__u32 *)(t2 + 1);
3326
+ return info1 == info2;
3327
+}
3328
+
3329
+/* Calculate type signature hash of ENUM. */
3330
+static long btf_hash_enum(struct btf_type *t)
3331
+{
3332
+ long h;
3333
+
3334
+ /* don't hash vlen and enum members to support enum fwd resolving */
3335
+ h = hash_combine(0, t->name_off);
3336
+ h = hash_combine(h, t->info & ~0xffff);
3337
+ h = hash_combine(h, t->size);
3338
+ return h;
3339
+}
3340
+
3341
+/* Check structural equality of two ENUMs. */
3342
+static bool btf_equal_enum(struct btf_type *t1, struct btf_type *t2)
3343
+{
3344
+ const struct btf_enum *m1, *m2;
3345
+ __u16 vlen;
3346
+ int i;
3347
+
3348
+ if (!btf_equal_common(t1, t2))
3349
+ return false;
3350
+
3351
+ vlen = btf_vlen(t1);
3352
+ m1 = btf_enum(t1);
3353
+ m2 = btf_enum(t2);
3354
+ for (i = 0; i < vlen; i++) {
3355
+ if (m1->name_off != m2->name_off || m1->val != m2->val)
3356
+ return false;
3357
+ m1++;
3358
+ m2++;
3359
+ }
3360
+ return true;
3361
+}
3362
+
3363
+static inline bool btf_is_enum_fwd(struct btf_type *t)
3364
+{
3365
+ return btf_is_enum(t) && btf_vlen(t) == 0;
3366
+}
3367
+
3368
+static bool btf_compat_enum(struct btf_type *t1, struct btf_type *t2)
3369
+{
3370
+ if (!btf_is_enum_fwd(t1) && !btf_is_enum_fwd(t2))
3371
+ return btf_equal_enum(t1, t2);
3372
+ /* ignore vlen when comparing */
3373
+ return t1->name_off == t2->name_off &&
3374
+ (t1->info & ~0xffff) == (t2->info & ~0xffff) &&
3375
+ t1->size == t2->size;
3376
+}
3377
+
3378
+/*
3379
+ * Calculate type signature hash of STRUCT/UNION, ignoring referenced type IDs,
3380
+ * as referenced type IDs equivalence is established separately during type
3381
+ * graph equivalence check algorithm.
3382
+ */
3383
+static long btf_hash_struct(struct btf_type *t)
3384
+{
3385
+ const struct btf_member *member = btf_members(t);
3386
+ __u32 vlen = btf_vlen(t);
3387
+ long h = btf_hash_common(t);
3388
+ int i;
3389
+
3390
+ for (i = 0; i < vlen; i++) {
3391
+ h = hash_combine(h, member->name_off);
3392
+ h = hash_combine(h, member->offset);
3393
+ /* no hashing of referenced type ID, it can be unresolved yet */
3394
+ member++;
3395
+ }
3396
+ return h;
3397
+}
3398
+
3399
+/*
3400
+ * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3401
+ * IDs. This check is performed during type graph equivalence check and
3402
+ * referenced types equivalence is checked separately.
3403
+ */
3404
+static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
3405
+{
3406
+ const struct btf_member *m1, *m2;
3407
+ __u16 vlen;
3408
+ int i;
3409
+
3410
+ if (!btf_equal_common(t1, t2))
3411
+ return false;
3412
+
3413
+ vlen = btf_vlen(t1);
3414
+ m1 = btf_members(t1);
3415
+ m2 = btf_members(t2);
3416
+ for (i = 0; i < vlen; i++) {
3417
+ if (m1->name_off != m2->name_off || m1->offset != m2->offset)
3418
+ return false;
3419
+ m1++;
3420
+ m2++;
3421
+ }
3422
+ return true;
3423
+}
3424
+
3425
+/*
3426
+ * Calculate type signature hash of ARRAY, including referenced type IDs,
3427
+ * under assumption that they were already resolved to canonical type IDs and
3428
+ * are not going to change.
3429
+ */
3430
+static long btf_hash_array(struct btf_type *t)
3431
+{
3432
+ const struct btf_array *info = btf_array(t);
3433
+ long h = btf_hash_common(t);
3434
+
3435
+ h = hash_combine(h, info->type);
3436
+ h = hash_combine(h, info->index_type);
3437
+ h = hash_combine(h, info->nelems);
3438
+ return h;
3439
+}
3440
+
3441
+/*
3442
+ * Check exact equality of two ARRAYs, taking into account referenced
3443
+ * type IDs, under assumption that they were already resolved to canonical
3444
+ * type IDs and are not going to change.
3445
+ * This function is called during reference types deduplication to compare
3446
+ * ARRAY to potential canonical representative.
3447
+ */
3448
+static bool btf_equal_array(struct btf_type *t1, struct btf_type *t2)
3449
+{
3450
+ const struct btf_array *info1, *info2;
3451
+
3452
+ if (!btf_equal_common(t1, t2))
3453
+ return false;
3454
+
3455
+ info1 = btf_array(t1);
3456
+ info2 = btf_array(t2);
3457
+ return info1->type == info2->type &&
3458
+ info1->index_type == info2->index_type &&
3459
+ info1->nelems == info2->nelems;
3460
+}
3461
+
3462
+/*
3463
+ * Check structural compatibility of two ARRAYs, ignoring referenced type
3464
+ * IDs. This check is performed during type graph equivalence check and
3465
+ * referenced types equivalence is checked separately.
3466
+ */
3467
+static bool btf_compat_array(struct btf_type *t1, struct btf_type *t2)
3468
+{
3469
+ if (!btf_equal_common(t1, t2))
3470
+ return false;
3471
+
3472
+ return btf_array(t1)->nelems == btf_array(t2)->nelems;
3473
+}
3474
+
3475
+/*
3476
+ * Calculate type signature hash of FUNC_PROTO, including referenced type IDs,
3477
+ * under assumption that they were already resolved to canonical type IDs and
3478
+ * are not going to change.
3479
+ */
3480
+static long btf_hash_fnproto(struct btf_type *t)
3481
+{
3482
+ const struct btf_param *member = btf_params(t);
3483
+ __u16 vlen = btf_vlen(t);
3484
+ long h = btf_hash_common(t);
3485
+ int i;
3486
+
3487
+ for (i = 0; i < vlen; i++) {
3488
+ h = hash_combine(h, member->name_off);
3489
+ h = hash_combine(h, member->type);
3490
+ member++;
3491
+ }
3492
+ return h;
3493
+}
3494
+
3495
+/*
3496
+ * Check exact equality of two FUNC_PROTOs, taking into account referenced
3497
+ * type IDs, under assumption that they were already resolved to canonical
3498
+ * type IDs and are not going to change.
3499
+ * This function is called during reference types deduplication to compare
3500
+ * FUNC_PROTO to potential canonical representative.
3501
+ */
3502
+static bool btf_equal_fnproto(struct btf_type *t1, struct btf_type *t2)
3503
+{
3504
+ const struct btf_param *m1, *m2;
3505
+ __u16 vlen;
3506
+ int i;
3507
+
3508
+ if (!btf_equal_common(t1, t2))
3509
+ return false;
3510
+
3511
+ vlen = btf_vlen(t1);
3512
+ m1 = btf_params(t1);
3513
+ m2 = btf_params(t2);
3514
+ for (i = 0; i < vlen; i++) {
3515
+ if (m1->name_off != m2->name_off || m1->type != m2->type)
3516
+ return false;
3517
+ m1++;
3518
+ m2++;
3519
+ }
3520
+ return true;
3521
+}
3522
+
3523
+/*
3524
+ * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
3525
+ * IDs. This check is performed during type graph equivalence check and
3526
+ * referenced types equivalence is checked separately.
3527
+ */
3528
+static bool btf_compat_fnproto(struct btf_type *t1, struct btf_type *t2)
3529
+{
3530
+ const struct btf_param *m1, *m2;
3531
+ __u16 vlen;
3532
+ int i;
3533
+
3534
+ /* skip return type ID */
3535
+ if (t1->name_off != t2->name_off || t1->info != t2->info)
3536
+ return false;
3537
+
3538
+ vlen = btf_vlen(t1);
3539
+ m1 = btf_params(t1);
3540
+ m2 = btf_params(t2);
3541
+ for (i = 0; i < vlen; i++) {
3542
+ if (m1->name_off != m2->name_off)
3543
+ return false;
3544
+ m1++;
3545
+ m2++;
3546
+ }
3547
+ return true;
3548
+}
3549
+
3550
+/*
3551
+ * Deduplicate primitive types, that can't reference other types, by calculating
3552
+ * their type signature hash and comparing them with any possible canonical
3553
+ * candidate. If no canonical candidate matches, type itself is marked as
3554
+ * canonical and is added into `btf_dedup->dedup_table` as another candidate.
3555
+ */
3556
+static int btf_dedup_prim_type(struct btf_dedup *d, __u32 type_id)
3557
+{
3558
+ struct btf_type *t = btf_type_by_id(d->btf, type_id);
3559
+ struct hashmap_entry *hash_entry;
3560
+ struct btf_type *cand;
3561
+ /* if we don't find equivalent type, then we are canonical */
3562
+ __u32 new_id = type_id;
3563
+ __u32 cand_id;
3564
+ long h;
3565
+
3566
+ switch (btf_kind(t)) {
3567
+ case BTF_KIND_CONST:
3568
+ case BTF_KIND_VOLATILE:
3569
+ case BTF_KIND_RESTRICT:
3570
+ case BTF_KIND_PTR:
3571
+ case BTF_KIND_TYPEDEF:
3572
+ case BTF_KIND_ARRAY:
3573
+ case BTF_KIND_STRUCT:
3574
+ case BTF_KIND_UNION:
3575
+ case BTF_KIND_FUNC:
3576
+ case BTF_KIND_FUNC_PROTO:
3577
+ case BTF_KIND_VAR:
3578
+ case BTF_KIND_DATASEC:
3579
+ return 0;
3580
+
3581
+ case BTF_KIND_INT:
3582
+ h = btf_hash_int(t);
3583
+ for_each_dedup_cand(d, hash_entry, h) {
3584
+ cand_id = (__u32)(long)hash_entry->value;
3585
+ cand = btf_type_by_id(d->btf, cand_id);
3586
+ if (btf_equal_int(t, cand)) {
3587
+ new_id = cand_id;
3588
+ break;
3589
+ }
3590
+ }
3591
+ break;
3592
+
3593
+ case BTF_KIND_ENUM:
3594
+ h = btf_hash_enum(t);
3595
+ for_each_dedup_cand(d, hash_entry, h) {
3596
+ cand_id = (__u32)(long)hash_entry->value;
3597
+ cand = btf_type_by_id(d->btf, cand_id);
3598
+ if (btf_equal_enum(t, cand)) {
3599
+ new_id = cand_id;
3600
+ break;
3601
+ }
3602
+ if (d->opts.dont_resolve_fwds)
3603
+ continue;
3604
+ if (btf_compat_enum(t, cand)) {
3605
+ if (btf_is_enum_fwd(t)) {
3606
+ /* resolve fwd to full enum */
3607
+ new_id = cand_id;
3608
+ break;
3609
+ }
3610
+ /* resolve canonical enum fwd to full enum */
3611
+ d->map[cand_id] = type_id;
3612
+ }
3613
+ }
3614
+ break;
3615
+
3616
+ case BTF_KIND_FWD:
3617
+ h = btf_hash_common(t);
3618
+ for_each_dedup_cand(d, hash_entry, h) {
3619
+ cand_id = (__u32)(long)hash_entry->value;
3620
+ cand = btf_type_by_id(d->btf, cand_id);
3621
+ if (btf_equal_common(t, cand)) {
3622
+ new_id = cand_id;
3623
+ break;
3624
+ }
3625
+ }
3626
+ break;
3627
+
3628
+ default:
3629
+ return -EINVAL;
3630
+ }
3631
+
3632
+ d->map[type_id] = new_id;
3633
+ if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
3634
+ return -ENOMEM;
3635
+
3636
+ return 0;
3637
+}
3638
+
3639
+static int btf_dedup_prim_types(struct btf_dedup *d)
3640
+{
3641
+ int i, err;
3642
+
3643
+ for (i = 1; i <= d->btf->nr_types; i++) {
3644
+ err = btf_dedup_prim_type(d, i);
3645
+ if (err)
3646
+ return err;
3647
+ }
3648
+ return 0;
3649
+}
3650
+
3651
+/*
3652
+ * Check whether type is already mapped into canonical one (could be to itself).
3653
+ */
3654
+static inline bool is_type_mapped(struct btf_dedup *d, uint32_t type_id)
3655
+{
3656
+ return d->map[type_id] <= BTF_MAX_NR_TYPES;
3657
+}
3658
+
3659
+/*
3660
+ * Resolve type ID into its canonical type ID, if any; otherwise return original
3661
+ * type ID. If type is FWD and is resolved into STRUCT/UNION already, follow
3662
+ * STRUCT/UNION link and resolve it into canonical type ID as well.
3663
+ */
3664
+static inline __u32 resolve_type_id(struct btf_dedup *d, __u32 type_id)
3665
+{
3666
+ while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3667
+ type_id = d->map[type_id];
3668
+ return type_id;
3669
+}
3670
+
3671
+/*
3672
+ * Resolve FWD to underlying STRUCT/UNION, if any; otherwise return original
3673
+ * type ID.
3674
+ */
3675
+static uint32_t resolve_fwd_id(struct btf_dedup *d, uint32_t type_id)
3676
+{
3677
+ __u32 orig_type_id = type_id;
3678
+
3679
+ if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3680
+ return type_id;
3681
+
3682
+ while (is_type_mapped(d, type_id) && d->map[type_id] != type_id)
3683
+ type_id = d->map[type_id];
3684
+
3685
+ if (!btf_is_fwd(btf__type_by_id(d->btf, type_id)))
3686
+ return type_id;
3687
+
3688
+ return orig_type_id;
3689
+}
3690
+
3691
+
3692
+static inline __u16 btf_fwd_kind(struct btf_type *t)
3693
+{
3694
+ return btf_kflag(t) ? BTF_KIND_UNION : BTF_KIND_STRUCT;
3695
+}
3696
+
3697
+/*
3698
+ * Check equivalence of BTF type graph formed by candidate struct/union (we'll
3699
+ * call it "candidate graph" in this description for brevity) to a type graph
3700
+ * formed by (potential) canonical struct/union ("canonical graph" for brevity
3701
+ * here, though keep in mind that not all types in canonical graph are
3702
+ * necessarily canonical representatives themselves, some of them might be
3703
+ * duplicates or its uniqueness might not have been established yet).
3704
+ * Returns:
3705
+ * - >0, if type graphs are equivalent;
3706
+ * - 0, if not equivalent;
3707
+ * - <0, on error.
3708
+ *
3709
+ * Algorithm performs side-by-side DFS traversal of both type graphs and checks
3710
+ * equivalence of BTF types at each step. If at any point BTF types in candidate
3711
+ * and canonical graphs are not compatible structurally, whole graphs are
3712
+ * incompatible. If types are structurally equivalent (i.e., all information
3713
+ * except referenced type IDs is exactly the same), a mapping from `canon_id` to
3714
+ * a `cand_id` is recored in hypothetical mapping (`btf_dedup->hypot_map`).
3715
+ * If a type references other types, then those referenced types are checked
3716
+ * for equivalence recursively.
3717
+ *
3718
+ * During DFS traversal, if we find that for current `canon_id` type we
3719
+ * already have some mapping in hypothetical map, we check for two possible
3720
+ * situations:
3721
+ * - `canon_id` is mapped to exactly the same type as `cand_id`. This will
3722
+ * happen when type graphs have cycles. In this case we assume those two
3723
+ * types are equivalent.
3724
+ * - `canon_id` is mapped to different type. This is contradiction in our
3725
+ * hypothetical mapping, because same graph in canonical graph corresponds
3726
+ * to two different types in candidate graph, which for equivalent type
3727
+ * graphs shouldn't happen. This condition terminates equivalence check
3728
+ * with negative result.
3729
+ *
3730
+ * If type graphs traversal exhausts types to check and find no contradiction,
3731
+ * then type graphs are equivalent.
3732
+ *
3733
+ * When checking types for equivalence, there is one special case: FWD types.
3734
+ * If FWD type resolution is allowed and one of the types (either from canonical
3735
+ * or candidate graph) is FWD and other is STRUCT/UNION (depending on FWD's kind
3736
+ * flag) and their names match, hypothetical mapping is updated to point from
3737
+ * FWD to STRUCT/UNION. If graphs will be determined as equivalent successfully,
3738
+ * this mapping will be used to record FWD -> STRUCT/UNION mapping permanently.
3739
+ *
3740
+ * Technically, this could lead to incorrect FWD to STRUCT/UNION resolution,
3741
+ * if there are two exactly named (or anonymous) structs/unions that are
3742
+ * compatible structurally, one of which has FWD field, while other is concrete
3743
+ * STRUCT/UNION, but according to C sources they are different structs/unions
3744
+ * that are referencing different types with the same name. This is extremely
3745
+ * unlikely to happen, but btf_dedup API allows to disable FWD resolution if
3746
+ * this logic is causing problems.
3747
+ *
3748
+ * Doing FWD resolution means that both candidate and/or canonical graphs can
3749
+ * consists of portions of the graph that come from multiple compilation units.
3750
+ * This is due to the fact that types within single compilation unit are always
3751
+ * deduplicated and FWDs are already resolved, if referenced struct/union
3752
+ * definiton is available. So, if we had unresolved FWD and found corresponding
3753
+ * STRUCT/UNION, they will be from different compilation units. This
3754
+ * consequently means that when we "link" FWD to corresponding STRUCT/UNION,
3755
+ * type graph will likely have at least two different BTF types that describe
3756
+ * same type (e.g., most probably there will be two different BTF types for the
3757
+ * same 'int' primitive type) and could even have "overlapping" parts of type
3758
+ * graph that describe same subset of types.
3759
+ *
3760
+ * This in turn means that our assumption that each type in canonical graph
3761
+ * must correspond to exactly one type in candidate graph might not hold
3762
+ * anymore and will make it harder to detect contradictions using hypothetical
3763
+ * map. To handle this problem, we allow to follow FWD -> STRUCT/UNION
3764
+ * resolution only in canonical graph. FWDs in candidate graphs are never
3765
+ * resolved. To see why it's OK, let's check all possible situations w.r.t. FWDs
3766
+ * that can occur:
3767
+ * - Both types in canonical and candidate graphs are FWDs. If they are
3768
+ * structurally equivalent, then they can either be both resolved to the
3769
+ * same STRUCT/UNION or not resolved at all. In both cases they are
3770
+ * equivalent and there is no need to resolve FWD on candidate side.
3771
+ * - Both types in canonical and candidate graphs are concrete STRUCT/UNION,
3772
+ * so nothing to resolve as well, algorithm will check equivalence anyway.
3773
+ * - Type in canonical graph is FWD, while type in candidate is concrete
3774
+ * STRUCT/UNION. In this case candidate graph comes from single compilation
3775
+ * unit, so there is exactly one BTF type for each unique C type. After
3776
+ * resolving FWD into STRUCT/UNION, there might be more than one BTF type
3777
+ * in canonical graph mapping to single BTF type in candidate graph, but
3778
+ * because hypothetical mapping maps from canonical to candidate types, it's
3779
+ * alright, and we still maintain the property of having single `canon_id`
3780
+ * mapping to single `cand_id` (there could be two different `canon_id`
3781
+ * mapped to the same `cand_id`, but it's not contradictory).
3782
+ * - Type in canonical graph is concrete STRUCT/UNION, while type in candidate
3783
+ * graph is FWD. In this case we are just going to check compatibility of
3784
+ * STRUCT/UNION and corresponding FWD, and if they are compatible, we'll
3785
+ * assume that whatever STRUCT/UNION FWD resolves to must be equivalent to
3786
+ * a concrete STRUCT/UNION from canonical graph. If the rest of type graphs
3787
+ * turn out equivalent, we'll re-resolve FWD to concrete STRUCT/UNION from
3788
+ * canonical graph.
3789
+ */
3790
+static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
3791
+ __u32 canon_id)
3792
+{
3793
+ struct btf_type *cand_type;
3794
+ struct btf_type *canon_type;
3795
+ __u32 hypot_type_id;
3796
+ __u16 cand_kind;
3797
+ __u16 canon_kind;
3798
+ int i, eq;
3799
+
3800
+ /* if both resolve to the same canonical, they must be equivalent */
3801
+ if (resolve_type_id(d, cand_id) == resolve_type_id(d, canon_id))
3802
+ return 1;
3803
+
3804
+ canon_id = resolve_fwd_id(d, canon_id);
3805
+
3806
+ hypot_type_id = d->hypot_map[canon_id];
3807
+ if (hypot_type_id <= BTF_MAX_NR_TYPES)
3808
+ return hypot_type_id == cand_id;
3809
+
3810
+ if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
3811
+ return -ENOMEM;
3812
+
3813
+ cand_type = btf_type_by_id(d->btf, cand_id);
3814
+ canon_type = btf_type_by_id(d->btf, canon_id);
3815
+ cand_kind = btf_kind(cand_type);
3816
+ canon_kind = btf_kind(canon_type);
3817
+
3818
+ if (cand_type->name_off != canon_type->name_off)
3819
+ return 0;
3820
+
3821
+ /* FWD <--> STRUCT/UNION equivalence check, if enabled */
3822
+ if (!d->opts.dont_resolve_fwds
3823
+ && (cand_kind == BTF_KIND_FWD || canon_kind == BTF_KIND_FWD)
3824
+ && cand_kind != canon_kind) {
3825
+ __u16 real_kind;
3826
+ __u16 fwd_kind;
3827
+
3828
+ if (cand_kind == BTF_KIND_FWD) {
3829
+ real_kind = canon_kind;
3830
+ fwd_kind = btf_fwd_kind(cand_type);
3831
+ } else {
3832
+ real_kind = cand_kind;
3833
+ fwd_kind = btf_fwd_kind(canon_type);
3834
+ }
3835
+ return fwd_kind == real_kind;
3836
+ }
3837
+
3838
+ if (cand_kind != canon_kind)
3839
+ return 0;
3840
+
3841
+ switch (cand_kind) {
3842
+ case BTF_KIND_INT:
3843
+ return btf_equal_int(cand_type, canon_type);
3844
+
3845
+ case BTF_KIND_ENUM:
3846
+ if (d->opts.dont_resolve_fwds)
3847
+ return btf_equal_enum(cand_type, canon_type);
3848
+ else
3849
+ return btf_compat_enum(cand_type, canon_type);
3850
+
3851
+ case BTF_KIND_FWD:
3852
+ return btf_equal_common(cand_type, canon_type);
3853
+
3854
+ case BTF_KIND_CONST:
3855
+ case BTF_KIND_VOLATILE:
3856
+ case BTF_KIND_RESTRICT:
3857
+ case BTF_KIND_PTR:
3858
+ case BTF_KIND_TYPEDEF:
3859
+ case BTF_KIND_FUNC:
3860
+ if (cand_type->info != canon_type->info)
3861
+ return 0;
3862
+ return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
3863
+
3864
+ case BTF_KIND_ARRAY: {
3865
+ const struct btf_array *cand_arr, *canon_arr;
3866
+
3867
+ if (!btf_compat_array(cand_type, canon_type))
3868
+ return 0;
3869
+ cand_arr = btf_array(cand_type);
3870
+ canon_arr = btf_array(canon_type);
3871
+ eq = btf_dedup_is_equiv(d,
3872
+ cand_arr->index_type, canon_arr->index_type);
3873
+ if (eq <= 0)
3874
+ return eq;
3875
+ return btf_dedup_is_equiv(d, cand_arr->type, canon_arr->type);
3876
+ }
3877
+
3878
+ case BTF_KIND_STRUCT:
3879
+ case BTF_KIND_UNION: {
3880
+ const struct btf_member *cand_m, *canon_m;
3881
+ __u16 vlen;
3882
+
3883
+ if (!btf_shallow_equal_struct(cand_type, canon_type))
3884
+ return 0;
3885
+ vlen = btf_vlen(cand_type);
3886
+ cand_m = btf_members(cand_type);
3887
+ canon_m = btf_members(canon_type);
3888
+ for (i = 0; i < vlen; i++) {
3889
+ eq = btf_dedup_is_equiv(d, cand_m->type, canon_m->type);
3890
+ if (eq <= 0)
3891
+ return eq;
3892
+ cand_m++;
3893
+ canon_m++;
3894
+ }
3895
+
3896
+ return 1;
3897
+ }
3898
+
3899
+ case BTF_KIND_FUNC_PROTO: {
3900
+ const struct btf_param *cand_p, *canon_p;
3901
+ __u16 vlen;
3902
+
3903
+ if (!btf_compat_fnproto(cand_type, canon_type))
3904
+ return 0;
3905
+ eq = btf_dedup_is_equiv(d, cand_type->type, canon_type->type);
3906
+ if (eq <= 0)
3907
+ return eq;
3908
+ vlen = btf_vlen(cand_type);
3909
+ cand_p = btf_params(cand_type);
3910
+ canon_p = btf_params(canon_type);
3911
+ for (i = 0; i < vlen; i++) {
3912
+ eq = btf_dedup_is_equiv(d, cand_p->type, canon_p->type);
3913
+ if (eq <= 0)
3914
+ return eq;
3915
+ cand_p++;
3916
+ canon_p++;
3917
+ }
3918
+ return 1;
3919
+ }
3920
+
3921
+ default:
3922
+ return -EINVAL;
3923
+ }
3924
+ return 0;
3925
+}
3926
+
3927
+/*
3928
+ * Use hypothetical mapping, produced by successful type graph equivalence
3929
+ * check, to augment existing struct/union canonical mapping, where possible.
3930
+ *
3931
+ * If BTF_KIND_FWD resolution is allowed, this mapping is also used to record
3932
+ * FWD -> STRUCT/UNION correspondence as well. FWD resolution is bidirectional:
3933
+ * it doesn't matter if FWD type was part of canonical graph or candidate one,
3934
+ * we are recording the mapping anyway. As opposed to carefulness required
3935
+ * for struct/union correspondence mapping (described below), for FWD resolution
3936
+ * it's not important, as by the time that FWD type (reference type) will be
3937
+ * deduplicated all structs/unions will be deduped already anyway.
3938
+ *
3939
+ * Recording STRUCT/UNION mapping is purely a performance optimization and is
3940
+ * not required for correctness. It needs to be done carefully to ensure that
3941
+ * struct/union from candidate's type graph is not mapped into corresponding
3942
+ * struct/union from canonical type graph that itself hasn't been resolved into
3943
+ * canonical representative. The only guarantee we have is that canonical
3944
+ * struct/union was determined as canonical and that won't change. But any
3945
+ * types referenced through that struct/union fields could have been not yet
3946
+ * resolved, so in case like that it's too early to establish any kind of
3947
+ * correspondence between structs/unions.
3948
+ *
3949
+ * No canonical correspondence is derived for primitive types (they are already
3950
+ * deduplicated completely already anyway) or reference types (they rely on
3951
+ * stability of struct/union canonical relationship for equivalence checks).
3952
+ */
3953
+static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
3954
+{
3955
+ __u32 cand_type_id, targ_type_id;
3956
+ __u16 t_kind, c_kind;
3957
+ __u32 t_id, c_id;
3958
+ int i;
3959
+
3960
+ for (i = 0; i < d->hypot_cnt; i++) {
3961
+ cand_type_id = d->hypot_list[i];
3962
+ targ_type_id = d->hypot_map[cand_type_id];
3963
+ t_id = resolve_type_id(d, targ_type_id);
3964
+ c_id = resolve_type_id(d, cand_type_id);
3965
+ t_kind = btf_kind(btf__type_by_id(d->btf, t_id));
3966
+ c_kind = btf_kind(btf__type_by_id(d->btf, c_id));
3967
+ /*
3968
+ * Resolve FWD into STRUCT/UNION.
3969
+ * It's ok to resolve FWD into STRUCT/UNION that's not yet
3970
+ * mapped to canonical representative (as opposed to
3971
+ * STRUCT/UNION <--> STRUCT/UNION mapping logic below), because
3972
+ * eventually that struct is going to be mapped and all resolved
3973
+ * FWDs will automatically resolve to correct canonical
3974
+ * representative. This will happen before ref type deduping,
3975
+ * which critically depends on stability of these mapping. This
3976
+ * stability is not a requirement for STRUCT/UNION equivalence
3977
+ * checks, though.
3978
+ */
3979
+ if (t_kind != BTF_KIND_FWD && c_kind == BTF_KIND_FWD)
3980
+ d->map[c_id] = t_id;
3981
+ else if (t_kind == BTF_KIND_FWD && c_kind != BTF_KIND_FWD)
3982
+ d->map[t_id] = c_id;
3983
+
3984
+ if ((t_kind == BTF_KIND_STRUCT || t_kind == BTF_KIND_UNION) &&
3985
+ c_kind != BTF_KIND_FWD &&
3986
+ is_type_mapped(d, c_id) &&
3987
+ !is_type_mapped(d, t_id)) {
3988
+ /*
3989
+ * as a perf optimization, we can map struct/union
3990
+ * that's part of type graph we just verified for
3991
+ * equivalence. We can do that for struct/union that has
3992
+ * canonical representative only, though.
3993
+ */
3994
+ d->map[t_id] = c_id;
3995
+ }
3996
+ }
3997
+}
3998
+
3999
+/*
4000
+ * Deduplicate struct/union types.
4001
+ *
4002
+ * For each struct/union type its type signature hash is calculated, taking
4003
+ * into account type's name, size, number, order and names of fields, but
4004
+ * ignoring type ID's referenced from fields, because they might not be deduped
4005
+ * completely until after reference types deduplication phase. This type hash
4006
+ * is used to iterate over all potential canonical types, sharing same hash.
4007
+ * For each canonical candidate we check whether type graphs that they form
4008
+ * (through referenced types in fields and so on) are equivalent using algorithm
4009
+ * implemented in `btf_dedup_is_equiv`. If such equivalence is found and
4010
+ * BTF_KIND_FWD resolution is allowed, then hypothetical mapping
4011
+ * (btf_dedup->hypot_map) produced by aforementioned type graph equivalence
4012
+ * algorithm is used to record FWD -> STRUCT/UNION mapping. It's also used to
4013
+ * potentially map other structs/unions to their canonical representatives,
4014
+ * if such relationship hasn't yet been established. This speeds up algorithm
4015
+ * by eliminating some of the duplicate work.
4016
+ *
4017
+ * If no matching canonical representative was found, struct/union is marked
4018
+ * as canonical for itself and is added into btf_dedup->dedup_table hash map
4019
+ * for further look ups.
4020
+ */
4021
+static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
4022
+{
4023
+ struct btf_type *cand_type, *t;
4024
+ struct hashmap_entry *hash_entry;
4025
+ /* if we don't find equivalent type, then we are canonical */
4026
+ __u32 new_id = type_id;
4027
+ __u16 kind;
4028
+ long h;
4029
+
4030
+ /* already deduped or is in process of deduping (loop detected) */
4031
+ if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4032
+ return 0;
4033
+
4034
+ t = btf_type_by_id(d->btf, type_id);
4035
+ kind = btf_kind(t);
4036
+
4037
+ if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
4038
+ return 0;
4039
+
4040
+ h = btf_hash_struct(t);
4041
+ for_each_dedup_cand(d, hash_entry, h) {
4042
+ __u32 cand_id = (__u32)(long)hash_entry->value;
4043
+ int eq;
4044
+
4045
+ /*
4046
+ * Even though btf_dedup_is_equiv() checks for
4047
+ * btf_shallow_equal_struct() internally when checking two
4048
+ * structs (unions) for equivalence, we need to guard here
4049
+ * from picking matching FWD type as a dedup candidate.
4050
+ * This can happen due to hash collision. In such case just
4051
+ * relying on btf_dedup_is_equiv() would lead to potentially
4052
+ * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
4053
+ * FWD and compatible STRUCT/UNION are considered equivalent.
4054
+ */
4055
+ cand_type = btf_type_by_id(d->btf, cand_id);
4056
+ if (!btf_shallow_equal_struct(t, cand_type))
4057
+ continue;
4058
+
4059
+ btf_dedup_clear_hypot_map(d);
4060
+ eq = btf_dedup_is_equiv(d, type_id, cand_id);
4061
+ if (eq < 0)
4062
+ return eq;
4063
+ if (!eq)
4064
+ continue;
4065
+ new_id = cand_id;
4066
+ btf_dedup_merge_hypot_map(d);
4067
+ break;
4068
+ }
4069
+
4070
+ d->map[type_id] = new_id;
4071
+ if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4072
+ return -ENOMEM;
4073
+
4074
+ return 0;
4075
+}
4076
+
4077
+static int btf_dedup_struct_types(struct btf_dedup *d)
4078
+{
4079
+ int i, err;
4080
+
4081
+ for (i = 1; i <= d->btf->nr_types; i++) {
4082
+ err = btf_dedup_struct_type(d, i);
4083
+ if (err)
4084
+ return err;
4085
+ }
4086
+ return 0;
4087
+}
4088
+
4089
+/*
4090
+ * Deduplicate reference type.
4091
+ *
4092
+ * Once all primitive and struct/union types got deduplicated, we can easily
4093
+ * deduplicate all other (reference) BTF types. This is done in two steps:
4094
+ *
4095
+ * 1. Resolve all referenced type IDs into their canonical type IDs. This
4096
+ * resolution can be done either immediately for primitive or struct/union types
4097
+ * (because they were deduped in previous two phases) or recursively for
4098
+ * reference types. Recursion will always terminate at either primitive or
4099
+ * struct/union type, at which point we can "unwind" chain of reference types
4100
+ * one by one. There is no danger of encountering cycles because in C type
4101
+ * system the only way to form type cycle is through struct/union, so any chain
4102
+ * of reference types, even those taking part in a type cycle, will inevitably
4103
+ * reach struct/union at some point.
4104
+ *
4105
+ * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
4106
+ * becomes "stable", in the sense that no further deduplication will cause
4107
+ * any changes to it. With that, it's now possible to calculate type's signature
4108
+ * hash (this time taking into account referenced type IDs) and loop over all
4109
+ * potential canonical representatives. If no match was found, current type
4110
+ * will become canonical representative of itself and will be added into
4111
+ * btf_dedup->dedup_table as another possible canonical representative.
4112
+ */
4113
+static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
4114
+{
4115
+ struct hashmap_entry *hash_entry;
4116
+ __u32 new_id = type_id, cand_id;
4117
+ struct btf_type *t, *cand;
4118
+ /* if we don't find equivalent type, then we are representative type */
4119
+ int ref_type_id;
4120
+ long h;
4121
+
4122
+ if (d->map[type_id] == BTF_IN_PROGRESS_ID)
4123
+ return -ELOOP;
4124
+ if (d->map[type_id] <= BTF_MAX_NR_TYPES)
4125
+ return resolve_type_id(d, type_id);
4126
+
4127
+ t = btf_type_by_id(d->btf, type_id);
4128
+ d->map[type_id] = BTF_IN_PROGRESS_ID;
4129
+
4130
+ switch (btf_kind(t)) {
4131
+ case BTF_KIND_CONST:
4132
+ case BTF_KIND_VOLATILE:
4133
+ case BTF_KIND_RESTRICT:
4134
+ case BTF_KIND_PTR:
4135
+ case BTF_KIND_TYPEDEF:
4136
+ case BTF_KIND_FUNC:
4137
+ ref_type_id = btf_dedup_ref_type(d, t->type);
4138
+ if (ref_type_id < 0)
4139
+ return ref_type_id;
4140
+ t->type = ref_type_id;
4141
+
4142
+ h = btf_hash_common(t);
4143
+ for_each_dedup_cand(d, hash_entry, h) {
4144
+ cand_id = (__u32)(long)hash_entry->value;
4145
+ cand = btf_type_by_id(d->btf, cand_id);
4146
+ if (btf_equal_common(t, cand)) {
4147
+ new_id = cand_id;
4148
+ break;
4149
+ }
4150
+ }
4151
+ break;
4152
+
4153
+ case BTF_KIND_ARRAY: {
4154
+ struct btf_array *info = btf_array(t);
4155
+
4156
+ ref_type_id = btf_dedup_ref_type(d, info->type);
4157
+ if (ref_type_id < 0)
4158
+ return ref_type_id;
4159
+ info->type = ref_type_id;
4160
+
4161
+ ref_type_id = btf_dedup_ref_type(d, info->index_type);
4162
+ if (ref_type_id < 0)
4163
+ return ref_type_id;
4164
+ info->index_type = ref_type_id;
4165
+
4166
+ h = btf_hash_array(t);
4167
+ for_each_dedup_cand(d, hash_entry, h) {
4168
+ cand_id = (__u32)(long)hash_entry->value;
4169
+ cand = btf_type_by_id(d->btf, cand_id);
4170
+ if (btf_equal_array(t, cand)) {
4171
+ new_id = cand_id;
4172
+ break;
4173
+ }
4174
+ }
4175
+ break;
4176
+ }
4177
+
4178
+ case BTF_KIND_FUNC_PROTO: {
4179
+ struct btf_param *param;
4180
+ __u16 vlen;
4181
+ int i;
4182
+
4183
+ ref_type_id = btf_dedup_ref_type(d, t->type);
4184
+ if (ref_type_id < 0)
4185
+ return ref_type_id;
4186
+ t->type = ref_type_id;
4187
+
4188
+ vlen = btf_vlen(t);
4189
+ param = btf_params(t);
4190
+ for (i = 0; i < vlen; i++) {
4191
+ ref_type_id = btf_dedup_ref_type(d, param->type);
4192
+ if (ref_type_id < 0)
4193
+ return ref_type_id;
4194
+ param->type = ref_type_id;
4195
+ param++;
4196
+ }
4197
+
4198
+ h = btf_hash_fnproto(t);
4199
+ for_each_dedup_cand(d, hash_entry, h) {
4200
+ cand_id = (__u32)(long)hash_entry->value;
4201
+ cand = btf_type_by_id(d->btf, cand_id);
4202
+ if (btf_equal_fnproto(t, cand)) {
4203
+ new_id = cand_id;
4204
+ break;
4205
+ }
4206
+ }
4207
+ break;
4208
+ }
4209
+
4210
+ default:
4211
+ return -EINVAL;
4212
+ }
4213
+
4214
+ d->map[type_id] = new_id;
4215
+ if (type_id == new_id && btf_dedup_table_add(d, h, type_id))
4216
+ return -ENOMEM;
4217
+
4218
+ return new_id;
4219
+}
4220
+
4221
+static int btf_dedup_ref_types(struct btf_dedup *d)
4222
+{
4223
+ int i, err;
4224
+
4225
+ for (i = 1; i <= d->btf->nr_types; i++) {
4226
+ err = btf_dedup_ref_type(d, i);
4227
+ if (err < 0)
4228
+ return err;
4229
+ }
4230
+ /* we won't need d->dedup_table anymore */
4231
+ hashmap__free(d->dedup_table);
4232
+ d->dedup_table = NULL;
4233
+ return 0;
4234
+}
4235
+
4236
+/*
4237
+ * Compact types.
4238
+ *
4239
+ * After we established for each type its corresponding canonical representative
4240
+ * type, we now can eliminate types that are not canonical and leave only
4241
+ * canonical ones layed out sequentially in memory by copying them over
4242
+ * duplicates. During compaction btf_dedup->hypot_map array is reused to store
4243
+ * a map from original type ID to a new compacted type ID, which will be used
4244
+ * during next phase to "fix up" type IDs, referenced from struct/union and
4245
+ * reference types.
4246
+ */
4247
+static int btf_dedup_compact_types(struct btf_dedup *d)
4248
+{
4249
+ __u32 *new_offs;
4250
+ __u32 next_type_id = 1;
4251
+ void *p;
4252
+ int i, len;
4253
+
4254
+ /* we are going to reuse hypot_map to store compaction remapping */
4255
+ d->hypot_map[0] = 0;
4256
+ for (i = 1; i <= d->btf->nr_types; i++)
4257
+ d->hypot_map[i] = BTF_UNPROCESSED_ID;
4258
+
4259
+ p = d->btf->types_data;
4260
+
4261
+ for (i = 1; i <= d->btf->nr_types; i++) {
4262
+ if (d->map[i] != i)
4263
+ continue;
4264
+
4265
+ len = btf_type_size(btf__type_by_id(d->btf, i));
4266
+ if (len < 0)
4267
+ return len;
4268
+
4269
+ memmove(p, btf__type_by_id(d->btf, i), len);
4270
+ d->hypot_map[i] = next_type_id;
4271
+ d->btf->type_offs[next_type_id] = p - d->btf->types_data;
4272
+ p += len;
4273
+ next_type_id++;
4274
+ }
4275
+
4276
+ /* shrink struct btf's internal types index and update btf_header */
4277
+ d->btf->nr_types = next_type_id - 1;
4278
+ d->btf->type_offs_cap = d->btf->nr_types + 1;
4279
+ d->btf->hdr->type_len = p - d->btf->types_data;
4280
+ new_offs = libbpf_reallocarray(d->btf->type_offs, d->btf->type_offs_cap,
4281
+ sizeof(*new_offs));
4282
+ if (!new_offs)
4283
+ return -ENOMEM;
4284
+ d->btf->type_offs = new_offs;
4285
+ d->btf->hdr->str_off = d->btf->hdr->type_len;
4286
+ d->btf->raw_size = d->btf->hdr->hdr_len + d->btf->hdr->type_len + d->btf->hdr->str_len;
4287
+ return 0;
4288
+}
4289
+
4290
+/*
4291
+ * Figure out final (deduplicated and compacted) type ID for provided original
4292
+ * `type_id` by first resolving it into corresponding canonical type ID and
4293
+ * then mapping it to a deduplicated type ID, stored in btf_dedup->hypot_map,
4294
+ * which is populated during compaction phase.
4295
+ */
4296
+static int btf_dedup_remap_type_id(struct btf_dedup *d, __u32 type_id)
4297
+{
4298
+ __u32 resolved_type_id, new_type_id;
4299
+
4300
+ resolved_type_id = resolve_type_id(d, type_id);
4301
+ new_type_id = d->hypot_map[resolved_type_id];
4302
+ if (new_type_id > BTF_MAX_NR_TYPES)
4303
+ return -EINVAL;
4304
+ return new_type_id;
4305
+}
4306
+
4307
+/*
4308
+ * Remap referenced type IDs into deduped type IDs.
4309
+ *
4310
+ * After BTF types are deduplicated and compacted, their final type IDs may
4311
+ * differ from original ones. The map from original to a corresponding
4312
+ * deduped type ID is stored in btf_dedup->hypot_map and is populated during
4313
+ * compaction phase. During remapping phase we are rewriting all type IDs
4314
+ * referenced from any BTF type (e.g., struct fields, func proto args, etc) to
4315
+ * their final deduped type IDs.
4316
+ */
4317
+static int btf_dedup_remap_type(struct btf_dedup *d, __u32 type_id)
4318
+{
4319
+ struct btf_type *t = btf_type_by_id(d->btf, type_id);
4320
+ int i, r;
4321
+
4322
+ switch (btf_kind(t)) {
4323
+ case BTF_KIND_INT:
4324
+ case BTF_KIND_ENUM:
4325
+ break;
4326
+
4327
+ case BTF_KIND_FWD:
4328
+ case BTF_KIND_CONST:
4329
+ case BTF_KIND_VOLATILE:
4330
+ case BTF_KIND_RESTRICT:
4331
+ case BTF_KIND_PTR:
4332
+ case BTF_KIND_TYPEDEF:
4333
+ case BTF_KIND_FUNC:
4334
+ case BTF_KIND_VAR:
4335
+ r = btf_dedup_remap_type_id(d, t->type);
4336
+ if (r < 0)
4337
+ return r;
4338
+ t->type = r;
4339
+ break;
4340
+
4341
+ case BTF_KIND_ARRAY: {
4342
+ struct btf_array *arr_info = btf_array(t);
4343
+
4344
+ r = btf_dedup_remap_type_id(d, arr_info->type);
4345
+ if (r < 0)
4346
+ return r;
4347
+ arr_info->type = r;
4348
+ r = btf_dedup_remap_type_id(d, arr_info->index_type);
4349
+ if (r < 0)
4350
+ return r;
4351
+ arr_info->index_type = r;
4352
+ break;
4353
+ }
4354
+
4355
+ case BTF_KIND_STRUCT:
4356
+ case BTF_KIND_UNION: {
4357
+ struct btf_member *member = btf_members(t);
4358
+ __u16 vlen = btf_vlen(t);
4359
+
4360
+ for (i = 0; i < vlen; i++) {
4361
+ r = btf_dedup_remap_type_id(d, member->type);
4362
+ if (r < 0)
4363
+ return r;
4364
+ member->type = r;
4365
+ member++;
4366
+ }
4367
+ break;
4368
+ }
4369
+
4370
+ case BTF_KIND_FUNC_PROTO: {
4371
+ struct btf_param *param = btf_params(t);
4372
+ __u16 vlen = btf_vlen(t);
4373
+
4374
+ r = btf_dedup_remap_type_id(d, t->type);
4375
+ if (r < 0)
4376
+ return r;
4377
+ t->type = r;
4378
+
4379
+ for (i = 0; i < vlen; i++) {
4380
+ r = btf_dedup_remap_type_id(d, param->type);
4381
+ if (r < 0)
4382
+ return r;
4383
+ param->type = r;
4384
+ param++;
4385
+ }
4386
+ break;
4387
+ }
4388
+
4389
+ case BTF_KIND_DATASEC: {
4390
+ struct btf_var_secinfo *var = btf_var_secinfos(t);
4391
+ __u16 vlen = btf_vlen(t);
4392
+
4393
+ for (i = 0; i < vlen; i++) {
4394
+ r = btf_dedup_remap_type_id(d, var->type);
4395
+ if (r < 0)
4396
+ return r;
4397
+ var->type = r;
4398
+ var++;
4399
+ }
4400
+ break;
4401
+ }
4402
+
4403
+ default:
4404
+ return -EINVAL;
4405
+ }
4406
+
4407
+ return 0;
4408
+}
4409
+
4410
+static int btf_dedup_remap_types(struct btf_dedup *d)
4411
+{
4412
+ int i, r;
4413
+
4414
+ for (i = 1; i <= d->btf->nr_types; i++) {
4415
+ r = btf_dedup_remap_type(d, i);
4416
+ if (r < 0)
4417
+ return r;
4418
+ }
4419
+ return 0;
4420
+}
4421
+
4422
+/*
4423
+ * Probe few well-known locations for vmlinux kernel image and try to load BTF
4424
+ * data out of it to use for target BTF.
4425
+ */
4426
+struct btf *libbpf_find_kernel_btf(void)
4427
+{
4428
+ struct {
4429
+ const char *path_fmt;
4430
+ bool raw_btf;
4431
+ } locations[] = {
4432
+ /* try canonical vmlinux BTF through sysfs first */
4433
+ { "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
4434
+ /* fall back to trying to find vmlinux ELF on disk otherwise */
4435
+ { "/boot/vmlinux-%1$s" },
4436
+ { "/lib/modules/%1$s/vmlinux-%1$s" },
4437
+ { "/lib/modules/%1$s/build/vmlinux" },
4438
+ { "/usr/lib/modules/%1$s/kernel/vmlinux" },
4439
+ { "/usr/lib/debug/boot/vmlinux-%1$s" },
4440
+ { "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
4441
+ { "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
4442
+ };
4443
+ char path[PATH_MAX + 1];
4444
+ struct utsname buf;
4445
+ struct btf *btf;
4446
+ int i;
4447
+
4448
+ uname(&buf);
4449
+
4450
+ for (i = 0; i < ARRAY_SIZE(locations); i++) {
4451
+ snprintf(path, PATH_MAX, locations[i].path_fmt, buf.release);
4452
+
4453
+ if (access(path, R_OK))
4454
+ continue;
4455
+
4456
+ if (locations[i].raw_btf)
4457
+ btf = btf__parse_raw(path);
4458
+ else
4459
+ btf = btf__parse_elf(path, NULL);
4460
+
4461
+ pr_debug("loading kernel BTF '%s': %ld\n",
4462
+ path, IS_ERR(btf) ? PTR_ERR(btf) : 0);
4463
+ if (IS_ERR(btf))
4464
+ continue;
4465
+
4466
+ return btf;
4467
+ }
4468
+
4469
+ pr_warn("failed to find valid kernel BTF\n");
4470
+ return ERR_PTR(-ESRCH);
4471
+}