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