1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2
3 /*
4 * BTF-to-C type converter.
5 *
6 * Copyright (c) 2019 Facebook
7 */
8
9 #include <stdbool.h>
10 #include <stddef.h>
11 #include <stdlib.h>
12 #include <string.h>
13 #include <ctype.h>
14 #include <endian.h>
15 #include <errno.h>
16 #include <limits.h>
17 #include <linux/err.h>
18 #include <linux/btf.h>
19 #include <linux/kernel.h>
20 #include "btf.h"
21 #include "hashmap.h"
22 #include "libbpf.h"
23 #include "libbpf_internal.h"
24
25 static const char PREFIXES[] = "\t\t\t\t\t\t\t\t\t\t\t\t\t";
26 static const size_t PREFIX_CNT = sizeof(PREFIXES) - 1;
27
pfx(int lvl)28 static const char *pfx(int lvl)
29 {
30 return lvl >= PREFIX_CNT ? PREFIXES : &PREFIXES[PREFIX_CNT - lvl];
31 }
32
33 enum btf_dump_type_order_state {
34 NOT_ORDERED,
35 ORDERING,
36 ORDERED,
37 };
38
39 enum btf_dump_type_emit_state {
40 NOT_EMITTED,
41 EMITTING,
42 EMITTED,
43 };
44
45 /* per-type auxiliary state */
46 struct btf_dump_type_aux_state {
47 /* topological sorting state */
48 enum btf_dump_type_order_state order_state: 2;
49 /* emitting state used to determine the need for forward declaration */
50 enum btf_dump_type_emit_state emit_state: 2;
51 /* whether forward declaration was already emitted */
52 __u8 fwd_emitted: 1;
53 /* whether unique non-duplicate name was already assigned */
54 __u8 name_resolved: 1;
55 /* whether type is referenced from any other type */
56 __u8 referenced: 1;
57 };
58
59 /* indent string length; one indent string is added for each indent level */
60 #define BTF_DATA_INDENT_STR_LEN 32
61
62 /*
63 * Common internal data for BTF type data dump operations.
64 */
65 struct btf_dump_data {
66 const void *data_end; /* end of valid data to show */
67 bool compact;
68 bool skip_names;
69 bool emit_zeroes;
70 __u8 indent_lvl; /* base indent level */
71 char indent_str[BTF_DATA_INDENT_STR_LEN];
72 /* below are used during iteration */
73 int depth;
74 bool is_array_member;
75 bool is_array_terminated;
76 bool is_array_char;
77 };
78
79 struct btf_dump {
80 const struct btf *btf;
81 btf_dump_printf_fn_t printf_fn;
82 void *cb_ctx;
83 int ptr_sz;
84 bool strip_mods;
85 bool skip_anon_defs;
86 int last_id;
87
88 /* per-type auxiliary state */
89 struct btf_dump_type_aux_state *type_states;
90 size_t type_states_cap;
91 /* per-type optional cached unique name, must be freed, if present */
92 const char **cached_names;
93 size_t cached_names_cap;
94
95 /* topo-sorted list of dependent type definitions */
96 __u32 *emit_queue;
97 int emit_queue_cap;
98 int emit_queue_cnt;
99
100 /*
101 * stack of type declarations (e.g., chain of modifiers, arrays,
102 * funcs, etc)
103 */
104 __u32 *decl_stack;
105 int decl_stack_cap;
106 int decl_stack_cnt;
107
108 /* maps struct/union/enum name to a number of name occurrences */
109 struct hashmap *type_names;
110 /*
111 * maps typedef identifiers and enum value names to a number of such
112 * name occurrences
113 */
114 struct hashmap *ident_names;
115 /*
116 * data for typed display; allocated if needed.
117 */
118 struct btf_dump_data *typed_dump;
119 };
120
str_hash_fn(long key,void * ctx)121 static size_t str_hash_fn(long key, void *ctx)
122 {
123 return str_hash((void *)key);
124 }
125
str_equal_fn(long a,long b,void * ctx)126 static bool str_equal_fn(long a, long b, void *ctx)
127 {
128 return strcmp((void *)a, (void *)b) == 0;
129 }
130
btf_name_of(const struct btf_dump * d,__u32 name_off)131 static const char *btf_name_of(const struct btf_dump *d, __u32 name_off)
132 {
133 return btf__name_by_offset(d->btf, name_off);
134 }
135
btf_dump_printf(const struct btf_dump * d,const char * fmt,...)136 static void btf_dump_printf(const struct btf_dump *d, const char *fmt, ...)
137 {
138 va_list args;
139
140 va_start(args, fmt);
141 d->printf_fn(d->cb_ctx, fmt, args);
142 va_end(args);
143 }
144
145 static int btf_dump_mark_referenced(struct btf_dump *d);
146 static int btf_dump_resize(struct btf_dump *d);
147
btf_dump__new(const struct btf * btf,btf_dump_printf_fn_t printf_fn,void * ctx,const struct btf_dump_opts * opts)148 struct btf_dump *btf_dump__new(const struct btf *btf,
149 btf_dump_printf_fn_t printf_fn,
150 void *ctx,
151 const struct btf_dump_opts *opts)
152 {
153 struct btf_dump *d;
154 int err;
155
156 if (!OPTS_VALID(opts, btf_dump_opts))
157 return libbpf_err_ptr(-EINVAL);
158
159 if (!printf_fn)
160 return libbpf_err_ptr(-EINVAL);
161
162 d = calloc(1, sizeof(struct btf_dump));
163 if (!d)
164 return libbpf_err_ptr(-ENOMEM);
165
166 d->btf = btf;
167 d->printf_fn = printf_fn;
168 d->cb_ctx = ctx;
169 d->ptr_sz = btf__pointer_size(btf) ? : sizeof(void *);
170
171 d->type_names = hashmap__new(str_hash_fn, str_equal_fn, NULL);
172 if (IS_ERR(d->type_names)) {
173 err = PTR_ERR(d->type_names);
174 d->type_names = NULL;
175 goto err;
176 }
177 d->ident_names = hashmap__new(str_hash_fn, str_equal_fn, NULL);
178 if (IS_ERR(d->ident_names)) {
179 err = PTR_ERR(d->ident_names);
180 d->ident_names = NULL;
181 goto err;
182 }
183
184 err = btf_dump_resize(d);
185 if (err)
186 goto err;
187
188 return d;
189 err:
190 btf_dump__free(d);
191 return libbpf_err_ptr(err);
192 }
193
btf_dump_resize(struct btf_dump * d)194 static int btf_dump_resize(struct btf_dump *d)
195 {
196 int err, last_id = btf__type_cnt(d->btf) - 1;
197
198 if (last_id <= d->last_id)
199 return 0;
200
201 if (libbpf_ensure_mem((void **)&d->type_states, &d->type_states_cap,
202 sizeof(*d->type_states), last_id + 1))
203 return -ENOMEM;
204 if (libbpf_ensure_mem((void **)&d->cached_names, &d->cached_names_cap,
205 sizeof(*d->cached_names), last_id + 1))
206 return -ENOMEM;
207
208 if (d->last_id == 0) {
209 /* VOID is special */
210 d->type_states[0].order_state = ORDERED;
211 d->type_states[0].emit_state = EMITTED;
212 }
213
214 /* eagerly determine referenced types for anon enums */
215 err = btf_dump_mark_referenced(d);
216 if (err)
217 return err;
218
219 d->last_id = last_id;
220 return 0;
221 }
222
btf_dump_free_names(struct hashmap * map)223 static void btf_dump_free_names(struct hashmap *map)
224 {
225 size_t bkt;
226 struct hashmap_entry *cur;
227
228 if (!map)
229 return;
230
231 hashmap__for_each_entry(map, cur, bkt)
232 free((void *)cur->pkey);
233
234 hashmap__free(map);
235 }
236
btf_dump__free(struct btf_dump * d)237 void btf_dump__free(struct btf_dump *d)
238 {
239 int i;
240
241 if (IS_ERR_OR_NULL(d))
242 return;
243
244 free(d->type_states);
245 if (d->cached_names) {
246 /* any set cached name is owned by us and should be freed */
247 for (i = 0; i <= d->last_id; i++) {
248 if (d->cached_names[i])
249 free((void *)d->cached_names[i]);
250 }
251 }
252 free(d->cached_names);
253 free(d->emit_queue);
254 free(d->decl_stack);
255 btf_dump_free_names(d->type_names);
256 btf_dump_free_names(d->ident_names);
257
258 free(d);
259 }
260
261 static int btf_dump_order_type(struct btf_dump *d, __u32 id, bool through_ptr);
262 static void btf_dump_emit_type(struct btf_dump *d, __u32 id, __u32 cont_id);
263
264 /*
265 * Dump BTF type in a compilable C syntax, including all the necessary
266 * dependent types, necessary for compilation. If some of the dependent types
267 * were already emitted as part of previous btf_dump__dump_type() invocation
268 * for another type, they won't be emitted again. This API allows callers to
269 * filter out BTF types according to user-defined criterias and emitted only
270 * minimal subset of types, necessary to compile everything. Full struct/union
271 * definitions will still be emitted, even if the only usage is through
272 * pointer and could be satisfied with just a forward declaration.
273 *
274 * Dumping is done in two high-level passes:
275 * 1. Topologically sort type definitions to satisfy C rules of compilation.
276 * 2. Emit type definitions in C syntax.
277 *
278 * Returns 0 on success; <0, otherwise.
279 */
btf_dump__dump_type(struct btf_dump * d,__u32 id)280 int btf_dump__dump_type(struct btf_dump *d, __u32 id)
281 {
282 int err, i;
283
284 if (id >= btf__type_cnt(d->btf))
285 return libbpf_err(-EINVAL);
286
287 err = btf_dump_resize(d);
288 if (err)
289 return libbpf_err(err);
290
291 d->emit_queue_cnt = 0;
292 err = btf_dump_order_type(d, id, false);
293 if (err < 0)
294 return libbpf_err(err);
295
296 for (i = 0; i < d->emit_queue_cnt; i++)
297 btf_dump_emit_type(d, d->emit_queue[i], 0 /*top-level*/);
298
299 return 0;
300 }
301
302 /*
303 * Mark all types that are referenced from any other type. This is used to
304 * determine top-level anonymous enums that need to be emitted as an
305 * independent type declarations.
306 * Anonymous enums come in two flavors: either embedded in a struct's field
307 * definition, in which case they have to be declared inline as part of field
308 * type declaration; or as a top-level anonymous enum, typically used for
309 * declaring global constants. It's impossible to distinguish between two
310 * without knowning whether given enum type was referenced from other type:
311 * top-level anonymous enum won't be referenced by anything, while embedded
312 * one will.
313 */
btf_dump_mark_referenced(struct btf_dump * d)314 static int btf_dump_mark_referenced(struct btf_dump *d)
315 {
316 int i, j, n = btf__type_cnt(d->btf);
317 const struct btf_type *t;
318 __u16 vlen;
319
320 for (i = d->last_id + 1; i < n; i++) {
321 t = btf__type_by_id(d->btf, i);
322 vlen = btf_vlen(t);
323
324 switch (btf_kind(t)) {
325 case BTF_KIND_INT:
326 case BTF_KIND_ENUM:
327 case BTF_KIND_ENUM64:
328 case BTF_KIND_FWD:
329 case BTF_KIND_FLOAT:
330 break;
331
332 case BTF_KIND_VOLATILE:
333 case BTF_KIND_CONST:
334 case BTF_KIND_RESTRICT:
335 case BTF_KIND_PTR:
336 case BTF_KIND_TYPEDEF:
337 case BTF_KIND_FUNC:
338 case BTF_KIND_VAR:
339 case BTF_KIND_DECL_TAG:
340 case BTF_KIND_TYPE_TAG:
341 d->type_states[t->type].referenced = 1;
342 break;
343
344 case BTF_KIND_ARRAY: {
345 const struct btf_array *a = btf_array(t);
346
347 d->type_states[a->index_type].referenced = 1;
348 d->type_states[a->type].referenced = 1;
349 break;
350 }
351 case BTF_KIND_STRUCT:
352 case BTF_KIND_UNION: {
353 const struct btf_member *m = btf_members(t);
354
355 for (j = 0; j < vlen; j++, m++)
356 d->type_states[m->type].referenced = 1;
357 break;
358 }
359 case BTF_KIND_FUNC_PROTO: {
360 const struct btf_param *p = btf_params(t);
361
362 for (j = 0; j < vlen; j++, p++)
363 d->type_states[p->type].referenced = 1;
364 break;
365 }
366 case BTF_KIND_DATASEC: {
367 const struct btf_var_secinfo *v = btf_var_secinfos(t);
368
369 for (j = 0; j < vlen; j++, v++)
370 d->type_states[v->type].referenced = 1;
371 break;
372 }
373 default:
374 return -EINVAL;
375 }
376 }
377 return 0;
378 }
379
btf_dump_add_emit_queue_id(struct btf_dump * d,__u32 id)380 static int btf_dump_add_emit_queue_id(struct btf_dump *d, __u32 id)
381 {
382 __u32 *new_queue;
383 size_t new_cap;
384
385 if (d->emit_queue_cnt >= d->emit_queue_cap) {
386 new_cap = max(16, d->emit_queue_cap * 3 / 2);
387 new_queue = libbpf_reallocarray(d->emit_queue, new_cap, sizeof(new_queue[0]));
388 if (!new_queue)
389 return -ENOMEM;
390 d->emit_queue = new_queue;
391 d->emit_queue_cap = new_cap;
392 }
393
394 d->emit_queue[d->emit_queue_cnt++] = id;
395 return 0;
396 }
397
398 /*
399 * Determine order of emitting dependent types and specified type to satisfy
400 * C compilation rules. This is done through topological sorting with an
401 * additional complication which comes from C rules. The main idea for C is
402 * that if some type is "embedded" into a struct/union, it's size needs to be
403 * known at the time of definition of containing type. E.g., for:
404 *
405 * struct A {};
406 * struct B { struct A x; }
407 *
408 * struct A *HAS* to be defined before struct B, because it's "embedded",
409 * i.e., it is part of struct B layout. But in the following case:
410 *
411 * struct A;
412 * struct B { struct A *x; }
413 * struct A {};
414 *
415 * it's enough to just have a forward declaration of struct A at the time of
416 * struct B definition, as struct B has a pointer to struct A, so the size of
417 * field x is known without knowing struct A size: it's sizeof(void *).
418 *
419 * Unfortunately, there are some trickier cases we need to handle, e.g.:
420 *
421 * struct A {}; // if this was forward-declaration: compilation error
422 * struct B {
423 * struct { // anonymous struct
424 * struct A y;
425 * } *x;
426 * };
427 *
428 * In this case, struct B's field x is a pointer, so it's size is known
429 * regardless of the size of (anonymous) struct it points to. But because this
430 * struct is anonymous and thus defined inline inside struct B, *and* it
431 * embeds struct A, compiler requires full definition of struct A to be known
432 * before struct B can be defined. This creates a transitive dependency
433 * between struct A and struct B. If struct A was forward-declared before
434 * struct B definition and fully defined after struct B definition, that would
435 * trigger compilation error.
436 *
437 * All this means that while we are doing topological sorting on BTF type
438 * graph, we need to determine relationships between different types (graph
439 * nodes):
440 * - weak link (relationship) between X and Y, if Y *CAN* be
441 * forward-declared at the point of X definition;
442 * - strong link, if Y *HAS* to be fully-defined before X can be defined.
443 *
444 * The rule is as follows. Given a chain of BTF types from X to Y, if there is
445 * BTF_KIND_PTR type in the chain and at least one non-anonymous type
446 * Z (excluding X, including Y), then link is weak. Otherwise, it's strong.
447 * Weak/strong relationship is determined recursively during DFS traversal and
448 * is returned as a result from btf_dump_order_type().
449 *
450 * btf_dump_order_type() is trying to avoid unnecessary forward declarations,
451 * but it is not guaranteeing that no extraneous forward declarations will be
452 * emitted.
453 *
454 * To avoid extra work, algorithm marks some of BTF types as ORDERED, when
455 * it's done with them, but not for all (e.g., VOLATILE, CONST, RESTRICT,
456 * ARRAY, FUNC_PROTO), as weak/strong semantics for those depends on the
457 * entire graph path, so depending where from one came to that BTF type, it
458 * might cause weak or strong ordering. For types like STRUCT/UNION/INT/ENUM,
459 * once they are processed, there is no need to do it again, so they are
460 * marked as ORDERED. We can mark PTR as ORDERED as well, as it semi-forces
461 * weak link, unless subsequent referenced STRUCT/UNION/ENUM is anonymous. But
462 * in any case, once those are processed, no need to do it again, as the
463 * result won't change.
464 *
465 * Returns:
466 * - 1, if type is part of strong link (so there is strong topological
467 * ordering requirements);
468 * - 0, if type is part of weak link (so can be satisfied through forward
469 * declaration);
470 * - <0, on error (e.g., unsatisfiable type loop detected).
471 */
btf_dump_order_type(struct btf_dump * d,__u32 id,bool through_ptr)472 static int btf_dump_order_type(struct btf_dump *d, __u32 id, bool through_ptr)
473 {
474 /*
475 * Order state is used to detect strong link cycles, but only for BTF
476 * kinds that are or could be an independent definition (i.e.,
477 * stand-alone fwd decl, enum, typedef, struct, union). Ptrs, arrays,
478 * func_protos, modifiers are just means to get to these definitions.
479 * Int/void don't need definitions, they are assumed to be always
480 * properly defined. We also ignore datasec, var, and funcs for now.
481 * So for all non-defining kinds, we never even set ordering state,
482 * for defining kinds we set ORDERING and subsequently ORDERED if it
483 * forms a strong link.
484 */
485 struct btf_dump_type_aux_state *tstate = &d->type_states[id];
486 const struct btf_type *t;
487 __u16 vlen;
488 int err, i;
489
490 /* return true, letting typedefs know that it's ok to be emitted */
491 if (tstate->order_state == ORDERED)
492 return 1;
493
494 t = btf__type_by_id(d->btf, id);
495
496 if (tstate->order_state == ORDERING) {
497 /* type loop, but resolvable through fwd declaration */
498 if (btf_is_composite(t) && through_ptr && t->name_off != 0)
499 return 0;
500 pr_warn("unsatisfiable type cycle, id:[%u]\n", id);
501 return -ELOOP;
502 }
503
504 switch (btf_kind(t)) {
505 case BTF_KIND_INT:
506 case BTF_KIND_FLOAT:
507 tstate->order_state = ORDERED;
508 return 0;
509
510 case BTF_KIND_PTR:
511 err = btf_dump_order_type(d, t->type, true);
512 tstate->order_state = ORDERED;
513 return err;
514
515 case BTF_KIND_ARRAY:
516 return btf_dump_order_type(d, btf_array(t)->type, false);
517
518 case BTF_KIND_STRUCT:
519 case BTF_KIND_UNION: {
520 const struct btf_member *m = btf_members(t);
521 /*
522 * struct/union is part of strong link, only if it's embedded
523 * (so no ptr in a path) or it's anonymous (so has to be
524 * defined inline, even if declared through ptr)
525 */
526 if (through_ptr && t->name_off != 0)
527 return 0;
528
529 tstate->order_state = ORDERING;
530
531 vlen = btf_vlen(t);
532 for (i = 0; i < vlen; i++, m++) {
533 err = btf_dump_order_type(d, m->type, false);
534 if (err < 0)
535 return err;
536 }
537
538 if (t->name_off != 0) {
539 err = btf_dump_add_emit_queue_id(d, id);
540 if (err < 0)
541 return err;
542 }
543
544 tstate->order_state = ORDERED;
545 return 1;
546 }
547 case BTF_KIND_ENUM:
548 case BTF_KIND_ENUM64:
549 case BTF_KIND_FWD:
550 /*
551 * non-anonymous or non-referenced enums are top-level
552 * declarations and should be emitted. Same logic can be
553 * applied to FWDs, it won't hurt anyways.
554 */
555 if (t->name_off != 0 || !tstate->referenced) {
556 err = btf_dump_add_emit_queue_id(d, id);
557 if (err)
558 return err;
559 }
560 tstate->order_state = ORDERED;
561 return 1;
562
563 case BTF_KIND_TYPEDEF: {
564 int is_strong;
565
566 is_strong = btf_dump_order_type(d, t->type, through_ptr);
567 if (is_strong < 0)
568 return is_strong;
569
570 /* typedef is similar to struct/union w.r.t. fwd-decls */
571 if (through_ptr && !is_strong)
572 return 0;
573
574 /* typedef is always a named definition */
575 err = btf_dump_add_emit_queue_id(d, id);
576 if (err)
577 return err;
578
579 d->type_states[id].order_state = ORDERED;
580 return 1;
581 }
582 case BTF_KIND_VOLATILE:
583 case BTF_KIND_CONST:
584 case BTF_KIND_RESTRICT:
585 case BTF_KIND_TYPE_TAG:
586 return btf_dump_order_type(d, t->type, through_ptr);
587
588 case BTF_KIND_FUNC_PROTO: {
589 const struct btf_param *p = btf_params(t);
590 bool is_strong;
591
592 err = btf_dump_order_type(d, t->type, through_ptr);
593 if (err < 0)
594 return err;
595 is_strong = err > 0;
596
597 vlen = btf_vlen(t);
598 for (i = 0; i < vlen; i++, p++) {
599 err = btf_dump_order_type(d, p->type, through_ptr);
600 if (err < 0)
601 return err;
602 if (err > 0)
603 is_strong = true;
604 }
605 return is_strong;
606 }
607 case BTF_KIND_FUNC:
608 case BTF_KIND_VAR:
609 case BTF_KIND_DATASEC:
610 case BTF_KIND_DECL_TAG:
611 d->type_states[id].order_state = ORDERED;
612 return 0;
613
614 default:
615 return -EINVAL;
616 }
617 }
618
619 static void btf_dump_emit_missing_aliases(struct btf_dump *d, __u32 id,
620 const struct btf_type *t);
621
622 static void btf_dump_emit_struct_fwd(struct btf_dump *d, __u32 id,
623 const struct btf_type *t);
624 static void btf_dump_emit_struct_def(struct btf_dump *d, __u32 id,
625 const struct btf_type *t, int lvl);
626
627 static void btf_dump_emit_enum_fwd(struct btf_dump *d, __u32 id,
628 const struct btf_type *t);
629 static void btf_dump_emit_enum_def(struct btf_dump *d, __u32 id,
630 const struct btf_type *t, int lvl);
631
632 static void btf_dump_emit_fwd_def(struct btf_dump *d, __u32 id,
633 const struct btf_type *t);
634
635 static void btf_dump_emit_typedef_def(struct btf_dump *d, __u32 id,
636 const struct btf_type *t, int lvl);
637
638 /* a local view into a shared stack */
639 struct id_stack {
640 const __u32 *ids;
641 int cnt;
642 };
643
644 static void btf_dump_emit_type_decl(struct btf_dump *d, __u32 id,
645 const char *fname, int lvl);
646 static void btf_dump_emit_type_chain(struct btf_dump *d,
647 struct id_stack *decl_stack,
648 const char *fname, int lvl);
649
650 static const char *btf_dump_type_name(struct btf_dump *d, __u32 id);
651 static const char *btf_dump_ident_name(struct btf_dump *d, __u32 id);
652 static size_t btf_dump_name_dups(struct btf_dump *d, struct hashmap *name_map,
653 const char *orig_name);
654
btf_dump_is_blacklisted(struct btf_dump * d,__u32 id)655 static bool btf_dump_is_blacklisted(struct btf_dump *d, __u32 id)
656 {
657 const struct btf_type *t = btf__type_by_id(d->btf, id);
658
659 /* __builtin_va_list is a compiler built-in, which causes compilation
660 * errors, when compiling w/ different compiler, then used to compile
661 * original code (e.g., GCC to compile kernel, Clang to use generated
662 * C header from BTF). As it is built-in, it should be already defined
663 * properly internally in compiler.
664 */
665 if (t->name_off == 0)
666 return false;
667 return strcmp(btf_name_of(d, t->name_off), "__builtin_va_list") == 0;
668 }
669
670 /*
671 * Emit C-syntax definitions of types from chains of BTF types.
672 *
673 * High-level handling of determining necessary forward declarations are handled
674 * by btf_dump_emit_type() itself, but all nitty-gritty details of emitting type
675 * declarations/definitions in C syntax are handled by a combo of
676 * btf_dump_emit_type_decl()/btf_dump_emit_type_chain() w/ delegation to
677 * corresponding btf_dump_emit_*_{def,fwd}() functions.
678 *
679 * We also keep track of "containing struct/union type ID" to determine when
680 * we reference it from inside and thus can avoid emitting unnecessary forward
681 * declaration.
682 *
683 * This algorithm is designed in such a way, that even if some error occurs
684 * (either technical, e.g., out of memory, or logical, i.e., malformed BTF
685 * that doesn't comply to C rules completely), algorithm will try to proceed
686 * and produce as much meaningful output as possible.
687 */
btf_dump_emit_type(struct btf_dump * d,__u32 id,__u32 cont_id)688 static void btf_dump_emit_type(struct btf_dump *d, __u32 id, __u32 cont_id)
689 {
690 struct btf_dump_type_aux_state *tstate = &d->type_states[id];
691 bool top_level_def = cont_id == 0;
692 const struct btf_type *t;
693 __u16 kind;
694
695 if (tstate->emit_state == EMITTED)
696 return;
697
698 t = btf__type_by_id(d->btf, id);
699 kind = btf_kind(t);
700
701 if (tstate->emit_state == EMITTING) {
702 if (tstate->fwd_emitted)
703 return;
704
705 switch (kind) {
706 case BTF_KIND_STRUCT:
707 case BTF_KIND_UNION:
708 /*
709 * if we are referencing a struct/union that we are
710 * part of - then no need for fwd declaration
711 */
712 if (id == cont_id)
713 return;
714 if (t->name_off == 0) {
715 pr_warn("anonymous struct/union loop, id:[%u]\n",
716 id);
717 return;
718 }
719 btf_dump_emit_struct_fwd(d, id, t);
720 btf_dump_printf(d, ";\n\n");
721 tstate->fwd_emitted = 1;
722 break;
723 case BTF_KIND_TYPEDEF:
724 /*
725 * for typedef fwd_emitted means typedef definition
726 * was emitted, but it can be used only for "weak"
727 * references through pointer only, not for embedding
728 */
729 if (!btf_dump_is_blacklisted(d, id)) {
730 btf_dump_emit_typedef_def(d, id, t, 0);
731 btf_dump_printf(d, ";\n\n");
732 }
733 tstate->fwd_emitted = 1;
734 break;
735 default:
736 break;
737 }
738
739 return;
740 }
741
742 switch (kind) {
743 case BTF_KIND_INT:
744 /* Emit type alias definitions if necessary */
745 btf_dump_emit_missing_aliases(d, id, t);
746
747 tstate->emit_state = EMITTED;
748 break;
749 case BTF_KIND_ENUM:
750 case BTF_KIND_ENUM64:
751 if (top_level_def) {
752 btf_dump_emit_enum_def(d, id, t, 0);
753 btf_dump_printf(d, ";\n\n");
754 }
755 tstate->emit_state = EMITTED;
756 break;
757 case BTF_KIND_PTR:
758 case BTF_KIND_VOLATILE:
759 case BTF_KIND_CONST:
760 case BTF_KIND_RESTRICT:
761 case BTF_KIND_TYPE_TAG:
762 btf_dump_emit_type(d, t->type, cont_id);
763 break;
764 case BTF_KIND_ARRAY:
765 btf_dump_emit_type(d, btf_array(t)->type, cont_id);
766 break;
767 case BTF_KIND_FWD:
768 btf_dump_emit_fwd_def(d, id, t);
769 btf_dump_printf(d, ";\n\n");
770 tstate->emit_state = EMITTED;
771 break;
772 case BTF_KIND_TYPEDEF:
773 tstate->emit_state = EMITTING;
774 btf_dump_emit_type(d, t->type, id);
775 /*
776 * typedef can server as both definition and forward
777 * declaration; at this stage someone depends on
778 * typedef as a forward declaration (refers to it
779 * through pointer), so unless we already did it,
780 * emit typedef as a forward declaration
781 */
782 if (!tstate->fwd_emitted && !btf_dump_is_blacklisted(d, id)) {
783 btf_dump_emit_typedef_def(d, id, t, 0);
784 btf_dump_printf(d, ";\n\n");
785 }
786 tstate->emit_state = EMITTED;
787 break;
788 case BTF_KIND_STRUCT:
789 case BTF_KIND_UNION:
790 tstate->emit_state = EMITTING;
791 /* if it's a top-level struct/union definition or struct/union
792 * is anonymous, then in C we'll be emitting all fields and
793 * their types (as opposed to just `struct X`), so we need to
794 * make sure that all types, referenced from struct/union
795 * members have necessary forward-declarations, where
796 * applicable
797 */
798 if (top_level_def || t->name_off == 0) {
799 const struct btf_member *m = btf_members(t);
800 __u16 vlen = btf_vlen(t);
801 int i, new_cont_id;
802
803 new_cont_id = t->name_off == 0 ? cont_id : id;
804 for (i = 0; i < vlen; i++, m++)
805 btf_dump_emit_type(d, m->type, new_cont_id);
806 } else if (!tstate->fwd_emitted && id != cont_id) {
807 btf_dump_emit_struct_fwd(d, id, t);
808 btf_dump_printf(d, ";\n\n");
809 tstate->fwd_emitted = 1;
810 }
811
812 if (top_level_def) {
813 btf_dump_emit_struct_def(d, id, t, 0);
814 btf_dump_printf(d, ";\n\n");
815 tstate->emit_state = EMITTED;
816 } else {
817 tstate->emit_state = NOT_EMITTED;
818 }
819 break;
820 case BTF_KIND_FUNC_PROTO: {
821 const struct btf_param *p = btf_params(t);
822 __u16 n = btf_vlen(t);
823 int i;
824
825 btf_dump_emit_type(d, t->type, cont_id);
826 for (i = 0; i < n; i++, p++)
827 btf_dump_emit_type(d, p->type, cont_id);
828
829 break;
830 }
831 default:
832 break;
833 }
834 }
835
btf_is_struct_packed(const struct btf * btf,__u32 id,const struct btf_type * t)836 static bool btf_is_struct_packed(const struct btf *btf, __u32 id,
837 const struct btf_type *t)
838 {
839 const struct btf_member *m;
840 int max_align = 1, align, i, bit_sz;
841 __u16 vlen;
842
843 m = btf_members(t);
844 vlen = btf_vlen(t);
845 /* all non-bitfield fields have to be naturally aligned */
846 for (i = 0; i < vlen; i++, m++) {
847 align = btf__align_of(btf, m->type);
848 bit_sz = btf_member_bitfield_size(t, i);
849 if (align && bit_sz == 0 && m->offset % (8 * align) != 0)
850 return true;
851 max_align = max(align, max_align);
852 }
853 /* size of a non-packed struct has to be a multiple of its alignment */
854 if (t->size % max_align != 0)
855 return true;
856 /*
857 * if original struct was marked as packed, but its layout is
858 * naturally aligned, we'll detect that it's not packed
859 */
860 return false;
861 }
862
btf_dump_emit_bit_padding(const struct btf_dump * d,int cur_off,int next_off,int next_align,bool in_bitfield,int lvl)863 static void btf_dump_emit_bit_padding(const struct btf_dump *d,
864 int cur_off, int next_off, int next_align,
865 bool in_bitfield, int lvl)
866 {
867 const struct {
868 const char *name;
869 int bits;
870 } pads[] = {
871 {"long", d->ptr_sz * 8}, {"int", 32}, {"short", 16}, {"char", 8}
872 };
873 int new_off, pad_bits, bits, i;
874 const char *pad_type;
875
876 if (cur_off >= next_off)
877 return; /* no gap */
878
879 /* For filling out padding we want to take advantage of
880 * natural alignment rules to minimize unnecessary explicit
881 * padding. First, we find the largest type (among long, int,
882 * short, or char) that can be used to force naturally aligned
883 * boundary. Once determined, we'll use such type to fill in
884 * the remaining padding gap. In some cases we can rely on
885 * compiler filling some gaps, but sometimes we need to force
886 * alignment to close natural alignment with markers like
887 * `long: 0` (this is always the case for bitfields). Note
888 * that even if struct itself has, let's say 4-byte alignment
889 * (i.e., it only uses up to int-aligned types), using `long:
890 * X;` explicit padding doesn't actually change struct's
891 * overall alignment requirements, but compiler does take into
892 * account that type's (long, in this example) natural
893 * alignment requirements when adding implicit padding. We use
894 * this fact heavily and don't worry about ruining correct
895 * struct alignment requirement.
896 */
897 for (i = 0; i < ARRAY_SIZE(pads); i++) {
898 pad_bits = pads[i].bits;
899 pad_type = pads[i].name;
900
901 new_off = roundup(cur_off, pad_bits);
902 if (new_off <= next_off)
903 break;
904 }
905
906 if (new_off > cur_off && new_off <= next_off) {
907 /* We need explicit `<type>: 0` aligning mark if next
908 * field is right on alignment offset and its
909 * alignment requirement is less strict than <type>'s
910 * alignment (so compiler won't naturally align to the
911 * offset we expect), or if subsequent `<type>: X`,
912 * will actually completely fit in the remaining hole,
913 * making compiler basically ignore `<type>: X`
914 * completely.
915 */
916 if (in_bitfield ||
917 (new_off == next_off && roundup(cur_off, next_align * 8) != new_off) ||
918 (new_off != next_off && next_off - new_off <= new_off - cur_off))
919 /* but for bitfields we'll emit explicit bit count */
920 btf_dump_printf(d, "\n%s%s: %d;", pfx(lvl), pad_type,
921 in_bitfield ? new_off - cur_off : 0);
922 cur_off = new_off;
923 }
924
925 /* Now we know we start at naturally aligned offset for a chosen
926 * padding type (long, int, short, or char), and so the rest is just
927 * a straightforward filling of remaining padding gap with full
928 * `<type>: sizeof(<type>);` markers, except for the last one, which
929 * might need smaller than sizeof(<type>) padding.
930 */
931 while (cur_off != next_off) {
932 bits = min(next_off - cur_off, pad_bits);
933 if (bits == pad_bits) {
934 btf_dump_printf(d, "\n%s%s: %d;", pfx(lvl), pad_type, pad_bits);
935 cur_off += bits;
936 continue;
937 }
938 /* For the remainder padding that doesn't cover entire
939 * pad_type bit length, we pick the smallest necessary type.
940 * This is pure aesthetics, we could have just used `long`,
941 * but having smallest necessary one communicates better the
942 * scale of the padding gap.
943 */
944 for (i = ARRAY_SIZE(pads) - 1; i >= 0; i--) {
945 pad_type = pads[i].name;
946 pad_bits = pads[i].bits;
947 if (pad_bits < bits)
948 continue;
949
950 btf_dump_printf(d, "\n%s%s: %d;", pfx(lvl), pad_type, bits);
951 cur_off += bits;
952 break;
953 }
954 }
955 }
956
btf_dump_emit_struct_fwd(struct btf_dump * d,__u32 id,const struct btf_type * t)957 static void btf_dump_emit_struct_fwd(struct btf_dump *d, __u32 id,
958 const struct btf_type *t)
959 {
960 btf_dump_printf(d, "%s%s%s",
961 btf_is_struct(t) ? "struct" : "union",
962 t->name_off ? " " : "",
963 btf_dump_type_name(d, id));
964 }
965
btf_dump_emit_struct_def(struct btf_dump * d,__u32 id,const struct btf_type * t,int lvl)966 static void btf_dump_emit_struct_def(struct btf_dump *d,
967 __u32 id,
968 const struct btf_type *t,
969 int lvl)
970 {
971 const struct btf_member *m = btf_members(t);
972 bool is_struct = btf_is_struct(t);
973 bool packed, prev_bitfield = false;
974 int align, i, off = 0;
975 __u16 vlen = btf_vlen(t);
976
977 align = btf__align_of(d->btf, id);
978 packed = is_struct ? btf_is_struct_packed(d->btf, id, t) : 0;
979
980 btf_dump_printf(d, "%s%s%s {",
981 is_struct ? "struct" : "union",
982 t->name_off ? " " : "",
983 btf_dump_type_name(d, id));
984
985 for (i = 0; i < vlen; i++, m++) {
986 const char *fname;
987 int m_off, m_sz, m_align;
988 bool in_bitfield;
989
990 fname = btf_name_of(d, m->name_off);
991 m_sz = btf_member_bitfield_size(t, i);
992 m_off = btf_member_bit_offset(t, i);
993 m_align = packed ? 1 : btf__align_of(d->btf, m->type);
994
995 in_bitfield = prev_bitfield && m_sz != 0;
996
997 btf_dump_emit_bit_padding(d, off, m_off, m_align, in_bitfield, lvl + 1);
998 btf_dump_printf(d, "\n%s", pfx(lvl + 1));
999 btf_dump_emit_type_decl(d, m->type, fname, lvl + 1);
1000
1001 if (m_sz) {
1002 btf_dump_printf(d, ": %d", m_sz);
1003 off = m_off + m_sz;
1004 prev_bitfield = true;
1005 } else {
1006 m_sz = max((__s64)0, btf__resolve_size(d->btf, m->type));
1007 off = m_off + m_sz * 8;
1008 prev_bitfield = false;
1009 }
1010
1011 btf_dump_printf(d, ";");
1012 }
1013
1014 /* pad at the end, if necessary */
1015 if (is_struct)
1016 btf_dump_emit_bit_padding(d, off, t->size * 8, align, false, lvl + 1);
1017
1018 /*
1019 * Keep `struct empty {}` on a single line,
1020 * only print newline when there are regular or padding fields.
1021 */
1022 if (vlen || t->size) {
1023 btf_dump_printf(d, "\n");
1024 btf_dump_printf(d, "%s}", pfx(lvl));
1025 } else {
1026 btf_dump_printf(d, "}");
1027 }
1028 if (packed)
1029 btf_dump_printf(d, " __attribute__((packed))");
1030 }
1031
1032 static const char *missing_base_types[][2] = {
1033 /*
1034 * GCC emits typedefs to its internal __PolyX_t types when compiling Arm
1035 * SIMD intrinsics. Alias them to standard base types.
1036 */
1037 { "__Poly8_t", "unsigned char" },
1038 { "__Poly16_t", "unsigned short" },
1039 { "__Poly64_t", "unsigned long long" },
1040 { "__Poly128_t", "unsigned __int128" },
1041 };
1042
btf_dump_emit_missing_aliases(struct btf_dump * d,__u32 id,const struct btf_type * t)1043 static void btf_dump_emit_missing_aliases(struct btf_dump *d, __u32 id,
1044 const struct btf_type *t)
1045 {
1046 const char *name = btf_dump_type_name(d, id);
1047 int i;
1048
1049 for (i = 0; i < ARRAY_SIZE(missing_base_types); i++) {
1050 if (strcmp(name, missing_base_types[i][0]) == 0) {
1051 btf_dump_printf(d, "typedef %s %s;\n\n",
1052 missing_base_types[i][1], name);
1053 break;
1054 }
1055 }
1056 }
1057
btf_dump_emit_enum_fwd(struct btf_dump * d,__u32 id,const struct btf_type * t)1058 static void btf_dump_emit_enum_fwd(struct btf_dump *d, __u32 id,
1059 const struct btf_type *t)
1060 {
1061 btf_dump_printf(d, "enum %s", btf_dump_type_name(d, id));
1062 }
1063
btf_dump_emit_enum32_val(struct btf_dump * d,const struct btf_type * t,int lvl,__u16 vlen)1064 static void btf_dump_emit_enum32_val(struct btf_dump *d,
1065 const struct btf_type *t,
1066 int lvl, __u16 vlen)
1067 {
1068 const struct btf_enum *v = btf_enum(t);
1069 bool is_signed = btf_kflag(t);
1070 const char *fmt_str;
1071 const char *name;
1072 size_t dup_cnt;
1073 int i;
1074
1075 for (i = 0; i < vlen; i++, v++) {
1076 name = btf_name_of(d, v->name_off);
1077 /* enumerators share namespace with typedef idents */
1078 dup_cnt = btf_dump_name_dups(d, d->ident_names, name);
1079 if (dup_cnt > 1) {
1080 fmt_str = is_signed ? "\n%s%s___%zd = %d," : "\n%s%s___%zd = %u,";
1081 btf_dump_printf(d, fmt_str, pfx(lvl + 1), name, dup_cnt, v->val);
1082 } else {
1083 fmt_str = is_signed ? "\n%s%s = %d," : "\n%s%s = %u,";
1084 btf_dump_printf(d, fmt_str, pfx(lvl + 1), name, v->val);
1085 }
1086 }
1087 }
1088
btf_dump_emit_enum64_val(struct btf_dump * d,const struct btf_type * t,int lvl,__u16 vlen)1089 static void btf_dump_emit_enum64_val(struct btf_dump *d,
1090 const struct btf_type *t,
1091 int lvl, __u16 vlen)
1092 {
1093 const struct btf_enum64 *v = btf_enum64(t);
1094 bool is_signed = btf_kflag(t);
1095 const char *fmt_str;
1096 const char *name;
1097 size_t dup_cnt;
1098 __u64 val;
1099 int i;
1100
1101 for (i = 0; i < vlen; i++, v++) {
1102 name = btf_name_of(d, v->name_off);
1103 dup_cnt = btf_dump_name_dups(d, d->ident_names, name);
1104 val = btf_enum64_value(v);
1105 if (dup_cnt > 1) {
1106 fmt_str = is_signed ? "\n%s%s___%zd = %lldLL,"
1107 : "\n%s%s___%zd = %lluULL,";
1108 btf_dump_printf(d, fmt_str,
1109 pfx(lvl + 1), name, dup_cnt,
1110 (unsigned long long)val);
1111 } else {
1112 fmt_str = is_signed ? "\n%s%s = %lldLL,"
1113 : "\n%s%s = %lluULL,";
1114 btf_dump_printf(d, fmt_str,
1115 pfx(lvl + 1), name,
1116 (unsigned long long)val);
1117 }
1118 }
1119 }
btf_dump_emit_enum_def(struct btf_dump * d,__u32 id,const struct btf_type * t,int lvl)1120 static void btf_dump_emit_enum_def(struct btf_dump *d, __u32 id,
1121 const struct btf_type *t,
1122 int lvl)
1123 {
1124 __u16 vlen = btf_vlen(t);
1125
1126 btf_dump_printf(d, "enum%s%s",
1127 t->name_off ? " " : "",
1128 btf_dump_type_name(d, id));
1129
1130 if (!vlen)
1131 return;
1132
1133 btf_dump_printf(d, " {");
1134 if (btf_is_enum(t))
1135 btf_dump_emit_enum32_val(d, t, lvl, vlen);
1136 else
1137 btf_dump_emit_enum64_val(d, t, lvl, vlen);
1138 btf_dump_printf(d, "\n%s}", pfx(lvl));
1139
1140 /* special case enums with special sizes */
1141 if (t->size == 1) {
1142 /* one-byte enums can be forced with mode(byte) attribute */
1143 btf_dump_printf(d, " __attribute__((mode(byte)))");
1144 } else if (t->size == 8 && d->ptr_sz == 8) {
1145 /* enum can be 8-byte sized if one of the enumerator values
1146 * doesn't fit in 32-bit integer, or by adding mode(word)
1147 * attribute (but probably only on 64-bit architectures); do
1148 * our best here to try to satisfy the contract without adding
1149 * unnecessary attributes
1150 */
1151 bool needs_word_mode;
1152
1153 if (btf_is_enum(t)) {
1154 /* enum can't represent 64-bit values, so we need word mode */
1155 needs_word_mode = true;
1156 } else {
1157 /* enum64 needs mode(word) if none of its values has
1158 * non-zero upper 32-bits (which means that all values
1159 * fit in 32-bit integers and won't cause compiler to
1160 * bump enum to be 64-bit naturally
1161 */
1162 int i;
1163
1164 needs_word_mode = true;
1165 for (i = 0; i < vlen; i++) {
1166 if (btf_enum64(t)[i].val_hi32 != 0) {
1167 needs_word_mode = false;
1168 break;
1169 }
1170 }
1171 }
1172 if (needs_word_mode)
1173 btf_dump_printf(d, " __attribute__((mode(word)))");
1174 }
1175
1176 }
1177
btf_dump_emit_fwd_def(struct btf_dump * d,__u32 id,const struct btf_type * t)1178 static void btf_dump_emit_fwd_def(struct btf_dump *d, __u32 id,
1179 const struct btf_type *t)
1180 {
1181 const char *name = btf_dump_type_name(d, id);
1182
1183 if (btf_kflag(t))
1184 btf_dump_printf(d, "union %s", name);
1185 else
1186 btf_dump_printf(d, "struct %s", name);
1187 }
1188
btf_dump_emit_typedef_def(struct btf_dump * d,__u32 id,const struct btf_type * t,int lvl)1189 static void btf_dump_emit_typedef_def(struct btf_dump *d, __u32 id,
1190 const struct btf_type *t, int lvl)
1191 {
1192 const char *name = btf_dump_ident_name(d, id);
1193
1194 /*
1195 * Old GCC versions are emitting invalid typedef for __gnuc_va_list
1196 * pointing to VOID. This generates warnings from btf_dump() and
1197 * results in uncompilable header file, so we are fixing it up here
1198 * with valid typedef into __builtin_va_list.
1199 */
1200 if (t->type == 0 && strcmp(name, "__gnuc_va_list") == 0) {
1201 btf_dump_printf(d, "typedef __builtin_va_list __gnuc_va_list");
1202 return;
1203 }
1204
1205 btf_dump_printf(d, "typedef ");
1206 btf_dump_emit_type_decl(d, t->type, name, lvl);
1207 }
1208
btf_dump_push_decl_stack_id(struct btf_dump * d,__u32 id)1209 static int btf_dump_push_decl_stack_id(struct btf_dump *d, __u32 id)
1210 {
1211 __u32 *new_stack;
1212 size_t new_cap;
1213
1214 if (d->decl_stack_cnt >= d->decl_stack_cap) {
1215 new_cap = max(16, d->decl_stack_cap * 3 / 2);
1216 new_stack = libbpf_reallocarray(d->decl_stack, new_cap, sizeof(new_stack[0]));
1217 if (!new_stack)
1218 return -ENOMEM;
1219 d->decl_stack = new_stack;
1220 d->decl_stack_cap = new_cap;
1221 }
1222
1223 d->decl_stack[d->decl_stack_cnt++] = id;
1224
1225 return 0;
1226 }
1227
1228 /*
1229 * Emit type declaration (e.g., field type declaration in a struct or argument
1230 * declaration in function prototype) in correct C syntax.
1231 *
1232 * For most types it's trivial, but there are few quirky type declaration
1233 * cases worth mentioning:
1234 * - function prototypes (especially nesting of function prototypes);
1235 * - arrays;
1236 * - const/volatile/restrict for pointers vs other types.
1237 *
1238 * For a good discussion of *PARSING* C syntax (as a human), see
1239 * Peter van der Linden's "Expert C Programming: Deep C Secrets",
1240 * Ch.3 "Unscrambling Declarations in C".
1241 *
1242 * It won't help with BTF to C conversion much, though, as it's an opposite
1243 * problem. So we came up with this algorithm in reverse to van der Linden's
1244 * parsing algorithm. It goes from structured BTF representation of type
1245 * declaration to a valid compilable C syntax.
1246 *
1247 * For instance, consider this C typedef:
1248 * typedef const int * const * arr[10] arr_t;
1249 * It will be represented in BTF with this chain of BTF types:
1250 * [typedef] -> [array] -> [ptr] -> [const] -> [ptr] -> [const] -> [int]
1251 *
1252 * Notice how [const] modifier always goes before type it modifies in BTF type
1253 * graph, but in C syntax, const/volatile/restrict modifiers are written to
1254 * the right of pointers, but to the left of other types. There are also other
1255 * quirks, like function pointers, arrays of them, functions returning other
1256 * functions, etc.
1257 *
1258 * We handle that by pushing all the types to a stack, until we hit "terminal"
1259 * type (int/enum/struct/union/fwd). Then depending on the kind of a type on
1260 * top of a stack, modifiers are handled differently. Array/function pointers
1261 * have also wildly different syntax and how nesting of them are done. See
1262 * code for authoritative definition.
1263 *
1264 * To avoid allocating new stack for each independent chain of BTF types, we
1265 * share one bigger stack, with each chain working only on its own local view
1266 * of a stack frame. Some care is required to "pop" stack frames after
1267 * processing type declaration chain.
1268 */
btf_dump__emit_type_decl(struct btf_dump * d,__u32 id,const struct btf_dump_emit_type_decl_opts * opts)1269 int btf_dump__emit_type_decl(struct btf_dump *d, __u32 id,
1270 const struct btf_dump_emit_type_decl_opts *opts)
1271 {
1272 const char *fname;
1273 int lvl, err;
1274
1275 if (!OPTS_VALID(opts, btf_dump_emit_type_decl_opts))
1276 return libbpf_err(-EINVAL);
1277
1278 err = btf_dump_resize(d);
1279 if (err)
1280 return libbpf_err(err);
1281
1282 fname = OPTS_GET(opts, field_name, "");
1283 lvl = OPTS_GET(opts, indent_level, 0);
1284 d->strip_mods = OPTS_GET(opts, strip_mods, false);
1285 btf_dump_emit_type_decl(d, id, fname, lvl);
1286 d->strip_mods = false;
1287 return 0;
1288 }
1289
btf_dump_emit_type_decl(struct btf_dump * d,__u32 id,const char * fname,int lvl)1290 static void btf_dump_emit_type_decl(struct btf_dump *d, __u32 id,
1291 const char *fname, int lvl)
1292 {
1293 struct id_stack decl_stack;
1294 const struct btf_type *t;
1295 int err, stack_start;
1296
1297 stack_start = d->decl_stack_cnt;
1298 for (;;) {
1299 t = btf__type_by_id(d->btf, id);
1300 if (d->strip_mods && btf_is_mod(t))
1301 goto skip_mod;
1302
1303 err = btf_dump_push_decl_stack_id(d, id);
1304 if (err < 0) {
1305 /*
1306 * if we don't have enough memory for entire type decl
1307 * chain, restore stack, emit warning, and try to
1308 * proceed nevertheless
1309 */
1310 pr_warn("not enough memory for decl stack:%d", err);
1311 d->decl_stack_cnt = stack_start;
1312 return;
1313 }
1314 skip_mod:
1315 /* VOID */
1316 if (id == 0)
1317 break;
1318
1319 switch (btf_kind(t)) {
1320 case BTF_KIND_PTR:
1321 case BTF_KIND_VOLATILE:
1322 case BTF_KIND_CONST:
1323 case BTF_KIND_RESTRICT:
1324 case BTF_KIND_FUNC_PROTO:
1325 case BTF_KIND_TYPE_TAG:
1326 id = t->type;
1327 break;
1328 case BTF_KIND_ARRAY:
1329 id = btf_array(t)->type;
1330 break;
1331 case BTF_KIND_INT:
1332 case BTF_KIND_ENUM:
1333 case BTF_KIND_ENUM64:
1334 case BTF_KIND_FWD:
1335 case BTF_KIND_STRUCT:
1336 case BTF_KIND_UNION:
1337 case BTF_KIND_TYPEDEF:
1338 case BTF_KIND_FLOAT:
1339 goto done;
1340 default:
1341 pr_warn("unexpected type in decl chain, kind:%u, id:[%u]\n",
1342 btf_kind(t), id);
1343 goto done;
1344 }
1345 }
1346 done:
1347 /*
1348 * We might be inside a chain of declarations (e.g., array of function
1349 * pointers returning anonymous (so inlined) structs, having another
1350 * array field). Each of those needs its own "stack frame" to handle
1351 * emitting of declarations. Those stack frames are non-overlapping
1352 * portions of shared btf_dump->decl_stack. To make it a bit nicer to
1353 * handle this set of nested stacks, we create a view corresponding to
1354 * our own "stack frame" and work with it as an independent stack.
1355 * We'll need to clean up after emit_type_chain() returns, though.
1356 */
1357 decl_stack.ids = d->decl_stack + stack_start;
1358 decl_stack.cnt = d->decl_stack_cnt - stack_start;
1359 btf_dump_emit_type_chain(d, &decl_stack, fname, lvl);
1360 /*
1361 * emit_type_chain() guarantees that it will pop its entire decl_stack
1362 * frame before returning. But it works with a read-only view into
1363 * decl_stack, so it doesn't actually pop anything from the
1364 * perspective of shared btf_dump->decl_stack, per se. We need to
1365 * reset decl_stack state to how it was before us to avoid it growing
1366 * all the time.
1367 */
1368 d->decl_stack_cnt = stack_start;
1369 }
1370
btf_dump_emit_mods(struct btf_dump * d,struct id_stack * decl_stack)1371 static void btf_dump_emit_mods(struct btf_dump *d, struct id_stack *decl_stack)
1372 {
1373 const struct btf_type *t;
1374 __u32 id;
1375
1376 while (decl_stack->cnt) {
1377 id = decl_stack->ids[decl_stack->cnt - 1];
1378 t = btf__type_by_id(d->btf, id);
1379
1380 switch (btf_kind(t)) {
1381 case BTF_KIND_VOLATILE:
1382 btf_dump_printf(d, "volatile ");
1383 break;
1384 case BTF_KIND_CONST:
1385 btf_dump_printf(d, "const ");
1386 break;
1387 case BTF_KIND_RESTRICT:
1388 btf_dump_printf(d, "restrict ");
1389 break;
1390 default:
1391 return;
1392 }
1393 decl_stack->cnt--;
1394 }
1395 }
1396
btf_dump_drop_mods(struct btf_dump * d,struct id_stack * decl_stack)1397 static void btf_dump_drop_mods(struct btf_dump *d, struct id_stack *decl_stack)
1398 {
1399 const struct btf_type *t;
1400 __u32 id;
1401
1402 while (decl_stack->cnt) {
1403 id = decl_stack->ids[decl_stack->cnt - 1];
1404 t = btf__type_by_id(d->btf, id);
1405 if (!btf_is_mod(t))
1406 return;
1407 decl_stack->cnt--;
1408 }
1409 }
1410
btf_dump_emit_name(const struct btf_dump * d,const char * name,bool last_was_ptr)1411 static void btf_dump_emit_name(const struct btf_dump *d,
1412 const char *name, bool last_was_ptr)
1413 {
1414 bool separate = name[0] && !last_was_ptr;
1415
1416 btf_dump_printf(d, "%s%s", separate ? " " : "", name);
1417 }
1418
btf_dump_emit_type_chain(struct btf_dump * d,struct id_stack * decls,const char * fname,int lvl)1419 static void btf_dump_emit_type_chain(struct btf_dump *d,
1420 struct id_stack *decls,
1421 const char *fname, int lvl)
1422 {
1423 /*
1424 * last_was_ptr is used to determine if we need to separate pointer
1425 * asterisk (*) from previous part of type signature with space, so
1426 * that we get `int ***`, instead of `int * * *`. We default to true
1427 * for cases where we have single pointer in a chain. E.g., in ptr ->
1428 * func_proto case. func_proto will start a new emit_type_chain call
1429 * with just ptr, which should be emitted as (*) or (*<fname>), so we
1430 * don't want to prepend space for that last pointer.
1431 */
1432 bool last_was_ptr = true;
1433 const struct btf_type *t;
1434 const char *name;
1435 __u16 kind;
1436 __u32 id;
1437
1438 while (decls->cnt) {
1439 id = decls->ids[--decls->cnt];
1440 if (id == 0) {
1441 /* VOID is a special snowflake */
1442 btf_dump_emit_mods(d, decls);
1443 btf_dump_printf(d, "void");
1444 last_was_ptr = false;
1445 continue;
1446 }
1447
1448 t = btf__type_by_id(d->btf, id);
1449 kind = btf_kind(t);
1450
1451 switch (kind) {
1452 case BTF_KIND_INT:
1453 case BTF_KIND_FLOAT:
1454 btf_dump_emit_mods(d, decls);
1455 name = btf_name_of(d, t->name_off);
1456 btf_dump_printf(d, "%s", name);
1457 break;
1458 case BTF_KIND_STRUCT:
1459 case BTF_KIND_UNION:
1460 btf_dump_emit_mods(d, decls);
1461 /* inline anonymous struct/union */
1462 if (t->name_off == 0 && !d->skip_anon_defs)
1463 btf_dump_emit_struct_def(d, id, t, lvl);
1464 else
1465 btf_dump_emit_struct_fwd(d, id, t);
1466 break;
1467 case BTF_KIND_ENUM:
1468 case BTF_KIND_ENUM64:
1469 btf_dump_emit_mods(d, decls);
1470 /* inline anonymous enum */
1471 if (t->name_off == 0 && !d->skip_anon_defs)
1472 btf_dump_emit_enum_def(d, id, t, lvl);
1473 else
1474 btf_dump_emit_enum_fwd(d, id, t);
1475 break;
1476 case BTF_KIND_FWD:
1477 btf_dump_emit_mods(d, decls);
1478 btf_dump_emit_fwd_def(d, id, t);
1479 break;
1480 case BTF_KIND_TYPEDEF:
1481 btf_dump_emit_mods(d, decls);
1482 btf_dump_printf(d, "%s", btf_dump_ident_name(d, id));
1483 break;
1484 case BTF_KIND_PTR:
1485 btf_dump_printf(d, "%s", last_was_ptr ? "*" : " *");
1486 break;
1487 case BTF_KIND_VOLATILE:
1488 btf_dump_printf(d, " volatile");
1489 break;
1490 case BTF_KIND_CONST:
1491 btf_dump_printf(d, " const");
1492 break;
1493 case BTF_KIND_RESTRICT:
1494 btf_dump_printf(d, " restrict");
1495 break;
1496 case BTF_KIND_TYPE_TAG:
1497 btf_dump_emit_mods(d, decls);
1498 name = btf_name_of(d, t->name_off);
1499 btf_dump_printf(d, " __attribute__((btf_type_tag(\"%s\")))", name);
1500 break;
1501 case BTF_KIND_ARRAY: {
1502 const struct btf_array *a = btf_array(t);
1503 const struct btf_type *next_t;
1504 __u32 next_id;
1505 bool multidim;
1506 /*
1507 * GCC has a bug
1508 * (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=8354)
1509 * which causes it to emit extra const/volatile
1510 * modifiers for an array, if array's element type has
1511 * const/volatile modifiers. Clang doesn't do that.
1512 * In general, it doesn't seem very meaningful to have
1513 * a const/volatile modifier for array, so we are
1514 * going to silently skip them here.
1515 */
1516 btf_dump_drop_mods(d, decls);
1517
1518 if (decls->cnt == 0) {
1519 btf_dump_emit_name(d, fname, last_was_ptr);
1520 btf_dump_printf(d, "[%u]", a->nelems);
1521 return;
1522 }
1523
1524 next_id = decls->ids[decls->cnt - 1];
1525 next_t = btf__type_by_id(d->btf, next_id);
1526 multidim = btf_is_array(next_t);
1527 /* we need space if we have named non-pointer */
1528 if (fname[0] && !last_was_ptr)
1529 btf_dump_printf(d, " ");
1530 /* no parentheses for multi-dimensional array */
1531 if (!multidim)
1532 btf_dump_printf(d, "(");
1533 btf_dump_emit_type_chain(d, decls, fname, lvl);
1534 if (!multidim)
1535 btf_dump_printf(d, ")");
1536 btf_dump_printf(d, "[%u]", a->nelems);
1537 return;
1538 }
1539 case BTF_KIND_FUNC_PROTO: {
1540 const struct btf_param *p = btf_params(t);
1541 __u16 vlen = btf_vlen(t);
1542 int i;
1543
1544 /*
1545 * GCC emits extra volatile qualifier for
1546 * __attribute__((noreturn)) function pointers. Clang
1547 * doesn't do it. It's a GCC quirk for backwards
1548 * compatibility with code written for GCC <2.5. So,
1549 * similarly to extra qualifiers for array, just drop
1550 * them, instead of handling them.
1551 */
1552 btf_dump_drop_mods(d, decls);
1553 if (decls->cnt) {
1554 btf_dump_printf(d, " (");
1555 btf_dump_emit_type_chain(d, decls, fname, lvl);
1556 btf_dump_printf(d, ")");
1557 } else {
1558 btf_dump_emit_name(d, fname, last_was_ptr);
1559 }
1560 btf_dump_printf(d, "(");
1561 /*
1562 * Clang for BPF target generates func_proto with no
1563 * args as a func_proto with a single void arg (e.g.,
1564 * `int (*f)(void)` vs just `int (*f)()`). We are
1565 * going to emit valid empty args (void) syntax for
1566 * such case. Similarly and conveniently, valid
1567 * no args case can be special-cased here as well.
1568 */
1569 if (vlen == 0 || (vlen == 1 && p->type == 0)) {
1570 btf_dump_printf(d, "void)");
1571 return;
1572 }
1573
1574 for (i = 0; i < vlen; i++, p++) {
1575 if (i > 0)
1576 btf_dump_printf(d, ", ");
1577
1578 /* last arg of type void is vararg */
1579 if (i == vlen - 1 && p->type == 0) {
1580 btf_dump_printf(d, "...");
1581 break;
1582 }
1583
1584 name = btf_name_of(d, p->name_off);
1585 btf_dump_emit_type_decl(d, p->type, name, lvl);
1586 }
1587
1588 btf_dump_printf(d, ")");
1589 return;
1590 }
1591 default:
1592 pr_warn("unexpected type in decl chain, kind:%u, id:[%u]\n",
1593 kind, id);
1594 return;
1595 }
1596
1597 last_was_ptr = kind == BTF_KIND_PTR;
1598 }
1599
1600 btf_dump_emit_name(d, fname, last_was_ptr);
1601 }
1602
1603 /* show type name as (type_name) */
btf_dump_emit_type_cast(struct btf_dump * d,__u32 id,bool top_level)1604 static void btf_dump_emit_type_cast(struct btf_dump *d, __u32 id,
1605 bool top_level)
1606 {
1607 const struct btf_type *t;
1608
1609 /* for array members, we don't bother emitting type name for each
1610 * member to avoid the redundancy of
1611 * .name = (char[4])[(char)'f',(char)'o',(char)'o',]
1612 */
1613 if (d->typed_dump->is_array_member)
1614 return;
1615
1616 /* avoid type name specification for variable/section; it will be done
1617 * for the associated variable value(s).
1618 */
1619 t = btf__type_by_id(d->btf, id);
1620 if (btf_is_var(t) || btf_is_datasec(t))
1621 return;
1622
1623 if (top_level)
1624 btf_dump_printf(d, "(");
1625
1626 d->skip_anon_defs = true;
1627 d->strip_mods = true;
1628 btf_dump_emit_type_decl(d, id, "", 0);
1629 d->strip_mods = false;
1630 d->skip_anon_defs = false;
1631
1632 if (top_level)
1633 btf_dump_printf(d, ")");
1634 }
1635
1636 /* return number of duplicates (occurrences) of a given name */
btf_dump_name_dups(struct btf_dump * d,struct hashmap * name_map,const char * orig_name)1637 static size_t btf_dump_name_dups(struct btf_dump *d, struct hashmap *name_map,
1638 const char *orig_name)
1639 {
1640 char *old_name, *new_name;
1641 size_t dup_cnt = 0;
1642 int err;
1643
1644 new_name = strdup(orig_name);
1645 if (!new_name)
1646 return 1;
1647
1648 (void)hashmap__find(name_map, orig_name, &dup_cnt);
1649 dup_cnt++;
1650
1651 err = hashmap__set(name_map, new_name, dup_cnt, &old_name, NULL);
1652 if (err)
1653 free(new_name);
1654
1655 free(old_name);
1656
1657 return dup_cnt;
1658 }
1659
btf_dump_resolve_name(struct btf_dump * d,__u32 id,struct hashmap * name_map)1660 static const char *btf_dump_resolve_name(struct btf_dump *d, __u32 id,
1661 struct hashmap *name_map)
1662 {
1663 struct btf_dump_type_aux_state *s = &d->type_states[id];
1664 const struct btf_type *t = btf__type_by_id(d->btf, id);
1665 const char *orig_name = btf_name_of(d, t->name_off);
1666 const char **cached_name = &d->cached_names[id];
1667 size_t dup_cnt;
1668
1669 if (t->name_off == 0)
1670 return "";
1671
1672 if (s->name_resolved)
1673 return *cached_name ? *cached_name : orig_name;
1674
1675 if (btf_is_fwd(t) || (btf_is_enum(t) && btf_vlen(t) == 0)) {
1676 s->name_resolved = 1;
1677 return orig_name;
1678 }
1679
1680 dup_cnt = btf_dump_name_dups(d, name_map, orig_name);
1681 if (dup_cnt > 1) {
1682 const size_t max_len = 256;
1683 char new_name[max_len];
1684
1685 snprintf(new_name, max_len, "%s___%zu", orig_name, dup_cnt);
1686 *cached_name = strdup(new_name);
1687 }
1688
1689 s->name_resolved = 1;
1690 return *cached_name ? *cached_name : orig_name;
1691 }
1692
btf_dump_type_name(struct btf_dump * d,__u32 id)1693 static const char *btf_dump_type_name(struct btf_dump *d, __u32 id)
1694 {
1695 return btf_dump_resolve_name(d, id, d->type_names);
1696 }
1697
btf_dump_ident_name(struct btf_dump * d,__u32 id)1698 static const char *btf_dump_ident_name(struct btf_dump *d, __u32 id)
1699 {
1700 return btf_dump_resolve_name(d, id, d->ident_names);
1701 }
1702
1703 static int btf_dump_dump_type_data(struct btf_dump *d,
1704 const char *fname,
1705 const struct btf_type *t,
1706 __u32 id,
1707 const void *data,
1708 __u8 bits_offset,
1709 __u8 bit_sz);
1710
btf_dump_data_newline(struct btf_dump * d)1711 static const char *btf_dump_data_newline(struct btf_dump *d)
1712 {
1713 return d->typed_dump->compact || d->typed_dump->depth == 0 ? "" : "\n";
1714 }
1715
btf_dump_data_delim(struct btf_dump * d)1716 static const char *btf_dump_data_delim(struct btf_dump *d)
1717 {
1718 return d->typed_dump->depth == 0 ? "" : ",";
1719 }
1720
btf_dump_data_pfx(struct btf_dump * d)1721 static void btf_dump_data_pfx(struct btf_dump *d)
1722 {
1723 int i, lvl = d->typed_dump->indent_lvl + d->typed_dump->depth;
1724
1725 if (d->typed_dump->compact)
1726 return;
1727
1728 for (i = 0; i < lvl; i++)
1729 btf_dump_printf(d, "%s", d->typed_dump->indent_str);
1730 }
1731
1732 /* A macro is used here as btf_type_value[s]() appends format specifiers
1733 * to the format specifier passed in; these do the work of appending
1734 * delimiters etc while the caller simply has to specify the type values
1735 * in the format specifier + value(s).
1736 */
1737 #define btf_dump_type_values(d, fmt, ...) \
1738 btf_dump_printf(d, fmt "%s%s", \
1739 ##__VA_ARGS__, \
1740 btf_dump_data_delim(d), \
1741 btf_dump_data_newline(d))
1742
btf_dump_unsupported_data(struct btf_dump * d,const struct btf_type * t,__u32 id)1743 static int btf_dump_unsupported_data(struct btf_dump *d,
1744 const struct btf_type *t,
1745 __u32 id)
1746 {
1747 btf_dump_printf(d, "<unsupported kind:%u>", btf_kind(t));
1748 return -ENOTSUP;
1749 }
1750
btf_dump_get_bitfield_value(struct btf_dump * d,const struct btf_type * t,const void * data,__u8 bits_offset,__u8 bit_sz,__u64 * value)1751 static int btf_dump_get_bitfield_value(struct btf_dump *d,
1752 const struct btf_type *t,
1753 const void *data,
1754 __u8 bits_offset,
1755 __u8 bit_sz,
1756 __u64 *value)
1757 {
1758 __u16 left_shift_bits, right_shift_bits;
1759 const __u8 *bytes = data;
1760 __u8 nr_copy_bits;
1761 __u64 num = 0;
1762 int i;
1763
1764 /* Maximum supported bitfield size is 64 bits */
1765 if (t->size > 8) {
1766 pr_warn("unexpected bitfield size %d\n", t->size);
1767 return -EINVAL;
1768 }
1769
1770 /* Bitfield value retrieval is done in two steps; first relevant bytes are
1771 * stored in num, then we left/right shift num to eliminate irrelevant bits.
1772 */
1773 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
1774 for (i = t->size - 1; i >= 0; i--)
1775 num = num * 256 + bytes[i];
1776 nr_copy_bits = bit_sz + bits_offset;
1777 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
1778 for (i = 0; i < t->size; i++)
1779 num = num * 256 + bytes[i];
1780 nr_copy_bits = t->size * 8 - bits_offset;
1781 #else
1782 # error "Unrecognized __BYTE_ORDER__"
1783 #endif
1784 left_shift_bits = 64 - nr_copy_bits;
1785 right_shift_bits = 64 - bit_sz;
1786
1787 *value = (num << left_shift_bits) >> right_shift_bits;
1788
1789 return 0;
1790 }
1791
btf_dump_bitfield_check_zero(struct btf_dump * d,const struct btf_type * t,const void * data,__u8 bits_offset,__u8 bit_sz)1792 static int btf_dump_bitfield_check_zero(struct btf_dump *d,
1793 const struct btf_type *t,
1794 const void *data,
1795 __u8 bits_offset,
1796 __u8 bit_sz)
1797 {
1798 __u64 check_num;
1799 int err;
1800
1801 err = btf_dump_get_bitfield_value(d, t, data, bits_offset, bit_sz, &check_num);
1802 if (err)
1803 return err;
1804 if (check_num == 0)
1805 return -ENODATA;
1806 return 0;
1807 }
1808
btf_dump_bitfield_data(struct btf_dump * d,const struct btf_type * t,const void * data,__u8 bits_offset,__u8 bit_sz)1809 static int btf_dump_bitfield_data(struct btf_dump *d,
1810 const struct btf_type *t,
1811 const void *data,
1812 __u8 bits_offset,
1813 __u8 bit_sz)
1814 {
1815 __u64 print_num;
1816 int err;
1817
1818 err = btf_dump_get_bitfield_value(d, t, data, bits_offset, bit_sz, &print_num);
1819 if (err)
1820 return err;
1821
1822 btf_dump_type_values(d, "0x%llx", (unsigned long long)print_num);
1823
1824 return 0;
1825 }
1826
1827 /* ints, floats and ptrs */
btf_dump_base_type_check_zero(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data)1828 static int btf_dump_base_type_check_zero(struct btf_dump *d,
1829 const struct btf_type *t,
1830 __u32 id,
1831 const void *data)
1832 {
1833 static __u8 bytecmp[16] = {};
1834 int nr_bytes;
1835
1836 /* For pointer types, pointer size is not defined on a per-type basis.
1837 * On dump creation however, we store the pointer size.
1838 */
1839 if (btf_kind(t) == BTF_KIND_PTR)
1840 nr_bytes = d->ptr_sz;
1841 else
1842 nr_bytes = t->size;
1843
1844 if (nr_bytes < 1 || nr_bytes > 16) {
1845 pr_warn("unexpected size %d for id [%u]\n", nr_bytes, id);
1846 return -EINVAL;
1847 }
1848
1849 if (memcmp(data, bytecmp, nr_bytes) == 0)
1850 return -ENODATA;
1851 return 0;
1852 }
1853
ptr_is_aligned(const struct btf * btf,__u32 type_id,const void * data)1854 static bool ptr_is_aligned(const struct btf *btf, __u32 type_id,
1855 const void *data)
1856 {
1857 int alignment = btf__align_of(btf, type_id);
1858
1859 if (alignment == 0)
1860 return false;
1861
1862 return ((uintptr_t)data) % alignment == 0;
1863 }
1864
btf_dump_int_data(struct btf_dump * d,const struct btf_type * t,__u32 type_id,const void * data,__u8 bits_offset)1865 static int btf_dump_int_data(struct btf_dump *d,
1866 const struct btf_type *t,
1867 __u32 type_id,
1868 const void *data,
1869 __u8 bits_offset)
1870 {
1871 __u8 encoding = btf_int_encoding(t);
1872 bool sign = encoding & BTF_INT_SIGNED;
1873 char buf[16] __attribute__((aligned(16)));
1874 int sz = t->size;
1875
1876 if (sz == 0 || sz > sizeof(buf)) {
1877 pr_warn("unexpected size %d for id [%u]\n", sz, type_id);
1878 return -EINVAL;
1879 }
1880
1881 /* handle packed int data - accesses of integers not aligned on
1882 * int boundaries can cause problems on some platforms.
1883 */
1884 if (!ptr_is_aligned(d->btf, type_id, data)) {
1885 memcpy(buf, data, sz);
1886 data = buf;
1887 }
1888
1889 switch (sz) {
1890 case 16: {
1891 const __u64 *ints = data;
1892 __u64 lsi, msi;
1893
1894 /* avoid use of __int128 as some 32-bit platforms do not
1895 * support it.
1896 */
1897 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
1898 lsi = ints[0];
1899 msi = ints[1];
1900 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
1901 lsi = ints[1];
1902 msi = ints[0];
1903 #else
1904 # error "Unrecognized __BYTE_ORDER__"
1905 #endif
1906 if (msi == 0)
1907 btf_dump_type_values(d, "0x%llx", (unsigned long long)lsi);
1908 else
1909 btf_dump_type_values(d, "0x%llx%016llx", (unsigned long long)msi,
1910 (unsigned long long)lsi);
1911 break;
1912 }
1913 case 8:
1914 if (sign)
1915 btf_dump_type_values(d, "%lld", *(long long *)data);
1916 else
1917 btf_dump_type_values(d, "%llu", *(unsigned long long *)data);
1918 break;
1919 case 4:
1920 if (sign)
1921 btf_dump_type_values(d, "%d", *(__s32 *)data);
1922 else
1923 btf_dump_type_values(d, "%u", *(__u32 *)data);
1924 break;
1925 case 2:
1926 if (sign)
1927 btf_dump_type_values(d, "%d", *(__s16 *)data);
1928 else
1929 btf_dump_type_values(d, "%u", *(__u16 *)data);
1930 break;
1931 case 1:
1932 if (d->typed_dump->is_array_char) {
1933 /* check for null terminator */
1934 if (d->typed_dump->is_array_terminated)
1935 break;
1936 if (*(char *)data == '\0') {
1937 d->typed_dump->is_array_terminated = true;
1938 break;
1939 }
1940 if (isprint(*(char *)data)) {
1941 btf_dump_type_values(d, "'%c'", *(char *)data);
1942 break;
1943 }
1944 }
1945 if (sign)
1946 btf_dump_type_values(d, "%d", *(__s8 *)data);
1947 else
1948 btf_dump_type_values(d, "%u", *(__u8 *)data);
1949 break;
1950 default:
1951 pr_warn("unexpected sz %d for id [%u]\n", sz, type_id);
1952 return -EINVAL;
1953 }
1954 return 0;
1955 }
1956
1957 union float_data {
1958 long double ld;
1959 double d;
1960 float f;
1961 };
1962
btf_dump_float_data(struct btf_dump * d,const struct btf_type * t,__u32 type_id,const void * data)1963 static int btf_dump_float_data(struct btf_dump *d,
1964 const struct btf_type *t,
1965 __u32 type_id,
1966 const void *data)
1967 {
1968 const union float_data *flp = data;
1969 union float_data fl;
1970 int sz = t->size;
1971
1972 /* handle unaligned data; copy to local union */
1973 if (!ptr_is_aligned(d->btf, type_id, data)) {
1974 memcpy(&fl, data, sz);
1975 flp = &fl;
1976 }
1977
1978 switch (sz) {
1979 case 16:
1980 btf_dump_type_values(d, "%Lf", flp->ld);
1981 break;
1982 case 8:
1983 btf_dump_type_values(d, "%lf", flp->d);
1984 break;
1985 case 4:
1986 btf_dump_type_values(d, "%f", flp->f);
1987 break;
1988 default:
1989 pr_warn("unexpected size %d for id [%u]\n", sz, type_id);
1990 return -EINVAL;
1991 }
1992 return 0;
1993 }
1994
btf_dump_var_data(struct btf_dump * d,const struct btf_type * v,__u32 id,const void * data)1995 static int btf_dump_var_data(struct btf_dump *d,
1996 const struct btf_type *v,
1997 __u32 id,
1998 const void *data)
1999 {
2000 enum btf_func_linkage linkage = btf_var(v)->linkage;
2001 const struct btf_type *t;
2002 const char *l;
2003 __u32 type_id;
2004
2005 switch (linkage) {
2006 case BTF_FUNC_STATIC:
2007 l = "static ";
2008 break;
2009 case BTF_FUNC_EXTERN:
2010 l = "extern ";
2011 break;
2012 case BTF_FUNC_GLOBAL:
2013 default:
2014 l = "";
2015 break;
2016 }
2017
2018 /* format of output here is [linkage] [type] [varname] = (type)value,
2019 * for example "static int cpu_profile_flip = (int)1"
2020 */
2021 btf_dump_printf(d, "%s", l);
2022 type_id = v->type;
2023 t = btf__type_by_id(d->btf, type_id);
2024 btf_dump_emit_type_cast(d, type_id, false);
2025 btf_dump_printf(d, " %s = ", btf_name_of(d, v->name_off));
2026 return btf_dump_dump_type_data(d, NULL, t, type_id, data, 0, 0);
2027 }
2028
btf_dump_array_data(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data)2029 static int btf_dump_array_data(struct btf_dump *d,
2030 const struct btf_type *t,
2031 __u32 id,
2032 const void *data)
2033 {
2034 const struct btf_array *array = btf_array(t);
2035 const struct btf_type *elem_type;
2036 __u32 i, elem_type_id;
2037 __s64 elem_size;
2038 bool is_array_member;
2039
2040 elem_type_id = array->type;
2041 elem_type = skip_mods_and_typedefs(d->btf, elem_type_id, NULL);
2042 elem_size = btf__resolve_size(d->btf, elem_type_id);
2043 if (elem_size <= 0) {
2044 pr_warn("unexpected elem size %zd for array type [%u]\n",
2045 (ssize_t)elem_size, id);
2046 return -EINVAL;
2047 }
2048
2049 if (btf_is_int(elem_type)) {
2050 /*
2051 * BTF_INT_CHAR encoding never seems to be set for
2052 * char arrays, so if size is 1 and element is
2053 * printable as a char, we'll do that.
2054 */
2055 if (elem_size == 1)
2056 d->typed_dump->is_array_char = true;
2057 }
2058
2059 /* note that we increment depth before calling btf_dump_print() below;
2060 * this is intentional. btf_dump_data_newline() will not print a
2061 * newline for depth 0 (since this leaves us with trailing newlines
2062 * at the end of typed display), so depth is incremented first.
2063 * For similar reasons, we decrement depth before showing the closing
2064 * parenthesis.
2065 */
2066 d->typed_dump->depth++;
2067 btf_dump_printf(d, "[%s", btf_dump_data_newline(d));
2068
2069 /* may be a multidimensional array, so store current "is array member"
2070 * status so we can restore it correctly later.
2071 */
2072 is_array_member = d->typed_dump->is_array_member;
2073 d->typed_dump->is_array_member = true;
2074 for (i = 0; i < array->nelems; i++, data += elem_size) {
2075 if (d->typed_dump->is_array_terminated)
2076 break;
2077 btf_dump_dump_type_data(d, NULL, elem_type, elem_type_id, data, 0, 0);
2078 }
2079 d->typed_dump->is_array_member = is_array_member;
2080 d->typed_dump->depth--;
2081 btf_dump_data_pfx(d);
2082 btf_dump_type_values(d, "]");
2083
2084 return 0;
2085 }
2086
btf_dump_struct_data(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data)2087 static int btf_dump_struct_data(struct btf_dump *d,
2088 const struct btf_type *t,
2089 __u32 id,
2090 const void *data)
2091 {
2092 const struct btf_member *m = btf_members(t);
2093 __u16 n = btf_vlen(t);
2094 int i, err = 0;
2095
2096 /* note that we increment depth before calling btf_dump_print() below;
2097 * this is intentional. btf_dump_data_newline() will not print a
2098 * newline for depth 0 (since this leaves us with trailing newlines
2099 * at the end of typed display), so depth is incremented first.
2100 * For similar reasons, we decrement depth before showing the closing
2101 * parenthesis.
2102 */
2103 d->typed_dump->depth++;
2104 btf_dump_printf(d, "{%s", btf_dump_data_newline(d));
2105
2106 for (i = 0; i < n; i++, m++) {
2107 const struct btf_type *mtype;
2108 const char *mname;
2109 __u32 moffset;
2110 __u8 bit_sz;
2111
2112 mtype = btf__type_by_id(d->btf, m->type);
2113 mname = btf_name_of(d, m->name_off);
2114 moffset = btf_member_bit_offset(t, i);
2115
2116 bit_sz = btf_member_bitfield_size(t, i);
2117 err = btf_dump_dump_type_data(d, mname, mtype, m->type, data + moffset / 8,
2118 moffset % 8, bit_sz);
2119 if (err < 0)
2120 return err;
2121 }
2122 d->typed_dump->depth--;
2123 btf_dump_data_pfx(d);
2124 btf_dump_type_values(d, "}");
2125 return err;
2126 }
2127
2128 union ptr_data {
2129 unsigned int p;
2130 unsigned long long lp;
2131 };
2132
btf_dump_ptr_data(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data)2133 static int btf_dump_ptr_data(struct btf_dump *d,
2134 const struct btf_type *t,
2135 __u32 id,
2136 const void *data)
2137 {
2138 if (ptr_is_aligned(d->btf, id, data) && d->ptr_sz == sizeof(void *)) {
2139 btf_dump_type_values(d, "%p", *(void **)data);
2140 } else {
2141 union ptr_data pt;
2142
2143 memcpy(&pt, data, d->ptr_sz);
2144 if (d->ptr_sz == 4)
2145 btf_dump_type_values(d, "0x%x", pt.p);
2146 else
2147 btf_dump_type_values(d, "0x%llx", pt.lp);
2148 }
2149 return 0;
2150 }
2151
btf_dump_get_enum_value(struct btf_dump * d,const struct btf_type * t,const void * data,__u32 id,__s64 * value)2152 static int btf_dump_get_enum_value(struct btf_dump *d,
2153 const struct btf_type *t,
2154 const void *data,
2155 __u32 id,
2156 __s64 *value)
2157 {
2158 bool is_signed = btf_kflag(t);
2159
2160 if (!ptr_is_aligned(d->btf, id, data)) {
2161 __u64 val;
2162 int err;
2163
2164 err = btf_dump_get_bitfield_value(d, t, data, 0, 0, &val);
2165 if (err)
2166 return err;
2167 *value = (__s64)val;
2168 return 0;
2169 }
2170
2171 switch (t->size) {
2172 case 8:
2173 *value = *(__s64 *)data;
2174 return 0;
2175 case 4:
2176 *value = is_signed ? (__s64)*(__s32 *)data : *(__u32 *)data;
2177 return 0;
2178 case 2:
2179 *value = is_signed ? *(__s16 *)data : *(__u16 *)data;
2180 return 0;
2181 case 1:
2182 *value = is_signed ? *(__s8 *)data : *(__u8 *)data;
2183 return 0;
2184 default:
2185 pr_warn("unexpected size %d for enum, id:[%u]\n", t->size, id);
2186 return -EINVAL;
2187 }
2188 }
2189
btf_dump_enum_data(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data)2190 static int btf_dump_enum_data(struct btf_dump *d,
2191 const struct btf_type *t,
2192 __u32 id,
2193 const void *data)
2194 {
2195 bool is_signed;
2196 __s64 value;
2197 int i, err;
2198
2199 err = btf_dump_get_enum_value(d, t, data, id, &value);
2200 if (err)
2201 return err;
2202
2203 is_signed = btf_kflag(t);
2204 if (btf_is_enum(t)) {
2205 const struct btf_enum *e;
2206
2207 for (i = 0, e = btf_enum(t); i < btf_vlen(t); i++, e++) {
2208 if (value != e->val)
2209 continue;
2210 btf_dump_type_values(d, "%s", btf_name_of(d, e->name_off));
2211 return 0;
2212 }
2213
2214 btf_dump_type_values(d, is_signed ? "%d" : "%u", value);
2215 } else {
2216 const struct btf_enum64 *e;
2217
2218 for (i = 0, e = btf_enum64(t); i < btf_vlen(t); i++, e++) {
2219 if (value != btf_enum64_value(e))
2220 continue;
2221 btf_dump_type_values(d, "%s", btf_name_of(d, e->name_off));
2222 return 0;
2223 }
2224
2225 btf_dump_type_values(d, is_signed ? "%lldLL" : "%lluULL",
2226 (unsigned long long)value);
2227 }
2228 return 0;
2229 }
2230
btf_dump_datasec_data(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data)2231 static int btf_dump_datasec_data(struct btf_dump *d,
2232 const struct btf_type *t,
2233 __u32 id,
2234 const void *data)
2235 {
2236 const struct btf_var_secinfo *vsi;
2237 const struct btf_type *var;
2238 __u32 i;
2239 int err;
2240
2241 btf_dump_type_values(d, "SEC(\"%s\") ", btf_name_of(d, t->name_off));
2242
2243 for (i = 0, vsi = btf_var_secinfos(t); i < btf_vlen(t); i++, vsi++) {
2244 var = btf__type_by_id(d->btf, vsi->type);
2245 err = btf_dump_dump_type_data(d, NULL, var, vsi->type, data + vsi->offset, 0, 0);
2246 if (err < 0)
2247 return err;
2248 btf_dump_printf(d, ";");
2249 }
2250 return 0;
2251 }
2252
2253 /* return size of type, or if base type overflows, return -E2BIG. */
btf_dump_type_data_check_overflow(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data,__u8 bits_offset,__u8 bit_sz)2254 static int btf_dump_type_data_check_overflow(struct btf_dump *d,
2255 const struct btf_type *t,
2256 __u32 id,
2257 const void *data,
2258 __u8 bits_offset,
2259 __u8 bit_sz)
2260 {
2261 __s64 size;
2262
2263 if (bit_sz) {
2264 /* bits_offset is at most 7. bit_sz is at most 128. */
2265 __u8 nr_bytes = (bits_offset + bit_sz + 7) / 8;
2266
2267 /* When bit_sz is non zero, it is called from
2268 * btf_dump_struct_data() where it only cares about
2269 * negative error value.
2270 * Return nr_bytes in success case to make it
2271 * consistent as the regular integer case below.
2272 */
2273 return data + nr_bytes > d->typed_dump->data_end ? -E2BIG : nr_bytes;
2274 }
2275
2276 size = btf__resolve_size(d->btf, id);
2277
2278 if (size < 0 || size >= INT_MAX) {
2279 pr_warn("unexpected size [%zu] for id [%u]\n",
2280 (size_t)size, id);
2281 return -EINVAL;
2282 }
2283
2284 /* Only do overflow checking for base types; we do not want to
2285 * avoid showing part of a struct, union or array, even if we
2286 * do not have enough data to show the full object. By
2287 * restricting overflow checking to base types we can ensure
2288 * that partial display succeeds, while avoiding overflowing
2289 * and using bogus data for display.
2290 */
2291 t = skip_mods_and_typedefs(d->btf, id, NULL);
2292 if (!t) {
2293 pr_warn("unexpected error skipping mods/typedefs for id [%u]\n",
2294 id);
2295 return -EINVAL;
2296 }
2297
2298 switch (btf_kind(t)) {
2299 case BTF_KIND_INT:
2300 case BTF_KIND_FLOAT:
2301 case BTF_KIND_PTR:
2302 case BTF_KIND_ENUM:
2303 case BTF_KIND_ENUM64:
2304 if (data + bits_offset / 8 + size > d->typed_dump->data_end)
2305 return -E2BIG;
2306 break;
2307 default:
2308 break;
2309 }
2310 return (int)size;
2311 }
2312
btf_dump_type_data_check_zero(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data,__u8 bits_offset,__u8 bit_sz)2313 static int btf_dump_type_data_check_zero(struct btf_dump *d,
2314 const struct btf_type *t,
2315 __u32 id,
2316 const void *data,
2317 __u8 bits_offset,
2318 __u8 bit_sz)
2319 {
2320 __s64 value;
2321 int i, err;
2322
2323 /* toplevel exceptions; we show zero values if
2324 * - we ask for them (emit_zeros)
2325 * - if we are at top-level so we see "struct empty { }"
2326 * - or if we are an array member and the array is non-empty and
2327 * not a char array; we don't want to be in a situation where we
2328 * have an integer array 0, 1, 0, 1 and only show non-zero values.
2329 * If the array contains zeroes only, or is a char array starting
2330 * with a '\0', the array-level check_zero() will prevent showing it;
2331 * we are concerned with determining zero value at the array member
2332 * level here.
2333 */
2334 if (d->typed_dump->emit_zeroes || d->typed_dump->depth == 0 ||
2335 (d->typed_dump->is_array_member &&
2336 !d->typed_dump->is_array_char))
2337 return 0;
2338
2339 t = skip_mods_and_typedefs(d->btf, id, NULL);
2340
2341 switch (btf_kind(t)) {
2342 case BTF_KIND_INT:
2343 if (bit_sz)
2344 return btf_dump_bitfield_check_zero(d, t, data, bits_offset, bit_sz);
2345 return btf_dump_base_type_check_zero(d, t, id, data);
2346 case BTF_KIND_FLOAT:
2347 case BTF_KIND_PTR:
2348 return btf_dump_base_type_check_zero(d, t, id, data);
2349 case BTF_KIND_ARRAY: {
2350 const struct btf_array *array = btf_array(t);
2351 const struct btf_type *elem_type;
2352 __u32 elem_type_id, elem_size;
2353 bool ischar;
2354
2355 elem_type_id = array->type;
2356 elem_size = btf__resolve_size(d->btf, elem_type_id);
2357 elem_type = skip_mods_and_typedefs(d->btf, elem_type_id, NULL);
2358
2359 ischar = btf_is_int(elem_type) && elem_size == 1;
2360
2361 /* check all elements; if _any_ element is nonzero, all
2362 * of array is displayed. We make an exception however
2363 * for char arrays where the first element is 0; these
2364 * are considered zeroed also, even if later elements are
2365 * non-zero because the string is terminated.
2366 */
2367 for (i = 0; i < array->nelems; i++) {
2368 if (i == 0 && ischar && *(char *)data == 0)
2369 return -ENODATA;
2370 err = btf_dump_type_data_check_zero(d, elem_type,
2371 elem_type_id,
2372 data +
2373 (i * elem_size),
2374 bits_offset, 0);
2375 if (err != -ENODATA)
2376 return err;
2377 }
2378 return -ENODATA;
2379 }
2380 case BTF_KIND_STRUCT:
2381 case BTF_KIND_UNION: {
2382 const struct btf_member *m = btf_members(t);
2383 __u16 n = btf_vlen(t);
2384
2385 /* if any struct/union member is non-zero, the struct/union
2386 * is considered non-zero and dumped.
2387 */
2388 for (i = 0; i < n; i++, m++) {
2389 const struct btf_type *mtype;
2390 __u32 moffset;
2391
2392 mtype = btf__type_by_id(d->btf, m->type);
2393 moffset = btf_member_bit_offset(t, i);
2394
2395 /* btf_int_bits() does not store member bitfield size;
2396 * bitfield size needs to be stored here so int display
2397 * of member can retrieve it.
2398 */
2399 bit_sz = btf_member_bitfield_size(t, i);
2400 err = btf_dump_type_data_check_zero(d, mtype, m->type, data + moffset / 8,
2401 moffset % 8, bit_sz);
2402 if (err != ENODATA)
2403 return err;
2404 }
2405 return -ENODATA;
2406 }
2407 case BTF_KIND_ENUM:
2408 case BTF_KIND_ENUM64:
2409 err = btf_dump_get_enum_value(d, t, data, id, &value);
2410 if (err)
2411 return err;
2412 if (value == 0)
2413 return -ENODATA;
2414 return 0;
2415 default:
2416 return 0;
2417 }
2418 }
2419
2420 /* returns size of data dumped, or error. */
btf_dump_dump_type_data(struct btf_dump * d,const char * fname,const struct btf_type * t,__u32 id,const void * data,__u8 bits_offset,__u8 bit_sz)2421 static int btf_dump_dump_type_data(struct btf_dump *d,
2422 const char *fname,
2423 const struct btf_type *t,
2424 __u32 id,
2425 const void *data,
2426 __u8 bits_offset,
2427 __u8 bit_sz)
2428 {
2429 int size, err = 0;
2430
2431 size = btf_dump_type_data_check_overflow(d, t, id, data, bits_offset, bit_sz);
2432 if (size < 0)
2433 return size;
2434 err = btf_dump_type_data_check_zero(d, t, id, data, bits_offset, bit_sz);
2435 if (err) {
2436 /* zeroed data is expected and not an error, so simply skip
2437 * dumping such data. Record other errors however.
2438 */
2439 if (err == -ENODATA)
2440 return size;
2441 return err;
2442 }
2443 btf_dump_data_pfx(d);
2444
2445 if (!d->typed_dump->skip_names) {
2446 if (fname && strlen(fname) > 0)
2447 btf_dump_printf(d, ".%s = ", fname);
2448 btf_dump_emit_type_cast(d, id, true);
2449 }
2450
2451 t = skip_mods_and_typedefs(d->btf, id, NULL);
2452
2453 switch (btf_kind(t)) {
2454 case BTF_KIND_UNKN:
2455 case BTF_KIND_FWD:
2456 case BTF_KIND_FUNC:
2457 case BTF_KIND_FUNC_PROTO:
2458 case BTF_KIND_DECL_TAG:
2459 err = btf_dump_unsupported_data(d, t, id);
2460 break;
2461 case BTF_KIND_INT:
2462 if (bit_sz)
2463 err = btf_dump_bitfield_data(d, t, data, bits_offset, bit_sz);
2464 else
2465 err = btf_dump_int_data(d, t, id, data, bits_offset);
2466 break;
2467 case BTF_KIND_FLOAT:
2468 err = btf_dump_float_data(d, t, id, data);
2469 break;
2470 case BTF_KIND_PTR:
2471 err = btf_dump_ptr_data(d, t, id, data);
2472 break;
2473 case BTF_KIND_ARRAY:
2474 err = btf_dump_array_data(d, t, id, data);
2475 break;
2476 case BTF_KIND_STRUCT:
2477 case BTF_KIND_UNION:
2478 err = btf_dump_struct_data(d, t, id, data);
2479 break;
2480 case BTF_KIND_ENUM:
2481 case BTF_KIND_ENUM64:
2482 /* handle bitfield and int enum values */
2483 if (bit_sz) {
2484 __u64 print_num;
2485 __s64 enum_val;
2486
2487 err = btf_dump_get_bitfield_value(d, t, data, bits_offset, bit_sz,
2488 &print_num);
2489 if (err)
2490 break;
2491 enum_val = (__s64)print_num;
2492 err = btf_dump_enum_data(d, t, id, &enum_val);
2493 } else
2494 err = btf_dump_enum_data(d, t, id, data);
2495 break;
2496 case BTF_KIND_VAR:
2497 err = btf_dump_var_data(d, t, id, data);
2498 break;
2499 case BTF_KIND_DATASEC:
2500 err = btf_dump_datasec_data(d, t, id, data);
2501 break;
2502 default:
2503 pr_warn("unexpected kind [%u] for id [%u]\n",
2504 BTF_INFO_KIND(t->info), id);
2505 return -EINVAL;
2506 }
2507 if (err < 0)
2508 return err;
2509 return size;
2510 }
2511
btf_dump__dump_type_data(struct btf_dump * d,__u32 id,const void * data,size_t data_sz,const struct btf_dump_type_data_opts * opts)2512 int btf_dump__dump_type_data(struct btf_dump *d, __u32 id,
2513 const void *data, size_t data_sz,
2514 const struct btf_dump_type_data_opts *opts)
2515 {
2516 struct btf_dump_data typed_dump = {};
2517 const struct btf_type *t;
2518 int ret;
2519
2520 if (!OPTS_VALID(opts, btf_dump_type_data_opts))
2521 return libbpf_err(-EINVAL);
2522
2523 t = btf__type_by_id(d->btf, id);
2524 if (!t)
2525 return libbpf_err(-ENOENT);
2526
2527 d->typed_dump = &typed_dump;
2528 d->typed_dump->data_end = data + data_sz;
2529 d->typed_dump->indent_lvl = OPTS_GET(opts, indent_level, 0);
2530
2531 /* default indent string is a tab */
2532 if (!OPTS_GET(opts, indent_str, NULL))
2533 d->typed_dump->indent_str[0] = '\t';
2534 else
2535 libbpf_strlcpy(d->typed_dump->indent_str, opts->indent_str,
2536 sizeof(d->typed_dump->indent_str));
2537
2538 d->typed_dump->compact = OPTS_GET(opts, compact, false);
2539 d->typed_dump->skip_names = OPTS_GET(opts, skip_names, false);
2540 d->typed_dump->emit_zeroes = OPTS_GET(opts, emit_zeroes, false);
2541
2542 ret = btf_dump_dump_type_data(d, NULL, t, id, data, 0, 0);
2543
2544 d->typed_dump = NULL;
2545
2546 return libbpf_err(ret);
2547 }
2548