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 #include "str_error.h"
25
26 static const char PREFIXES[] = "\t\t\t\t\t\t\t\t\t\t\t\t\t";
27 static const size_t PREFIX_CNT = sizeof(PREFIXES) - 1;
28
pfx(int lvl)29 static const char *pfx(int lvl)
30 {
31 return lvl >= PREFIX_CNT ? PREFIXES : &PREFIXES[PREFIX_CNT - lvl];
32 }
33
34 enum btf_dump_type_order_state {
35 NOT_ORDERED,
36 ORDERING,
37 ORDERED,
38 };
39
40 enum btf_dump_type_emit_state {
41 NOT_EMITTED,
42 EMITTING,
43 EMITTED,
44 };
45
46 /* per-type auxiliary state */
47 struct btf_dump_type_aux_state {
48 /* topological sorting state */
49 enum btf_dump_type_order_state order_state: 2;
50 /* emitting state used to determine the need for forward declaration */
51 enum btf_dump_type_emit_state emit_state: 2;
52 /* whether forward declaration was already emitted */
53 __u8 fwd_emitted: 1;
54 /* whether unique non-duplicate name was already assigned */
55 __u8 name_resolved: 1;
56 /* whether type is referenced from any other type */
57 __u8 referenced: 1;
58 };
59
60 /* indent string length; one indent string is added for each indent level */
61 #define BTF_DATA_INDENT_STR_LEN 32
62
63 /*
64 * Common internal data for BTF type data dump operations.
65 */
66 struct btf_dump_data {
67 const void *data_end; /* end of valid data to show */
68 bool compact;
69 bool skip_names;
70 bool emit_zeroes;
71 __u8 indent_lvl; /* base indent level */
72 char indent_str[BTF_DATA_INDENT_STR_LEN];
73 /* below are used during iteration */
74 int depth;
75 bool is_array_member;
76 bool is_array_terminated;
77 bool is_array_char;
78 };
79
80 struct btf_dump {
81 const struct btf *btf;
82 btf_dump_printf_fn_t printf_fn;
83 void *cb_ctx;
84 int ptr_sz;
85 bool strip_mods;
86 bool skip_anon_defs;
87 int last_id;
88
89 /* per-type auxiliary state */
90 struct btf_dump_type_aux_state *type_states;
91 size_t type_states_cap;
92 /* per-type optional cached unique name, must be freed, if present */
93 const char **cached_names;
94 size_t cached_names_cap;
95
96 /* topo-sorted list of dependent type definitions */
97 __u32 *emit_queue;
98 int emit_queue_cap;
99 int emit_queue_cnt;
100
101 /*
102 * stack of type declarations (e.g., chain of modifiers, arrays,
103 * funcs, etc)
104 */
105 __u32 *decl_stack;
106 int decl_stack_cap;
107 int decl_stack_cnt;
108
109 /* maps struct/union/enum name to a number of name occurrences */
110 struct hashmap *type_names;
111 /*
112 * maps typedef identifiers and enum value names to a number of such
113 * name occurrences
114 */
115 struct hashmap *ident_names;
116 /*
117 * data for typed display; allocated if needed.
118 */
119 struct btf_dump_data *typed_dump;
120 };
121
str_hash_fn(long key,void * ctx)122 static size_t str_hash_fn(long key, void *ctx)
123 {
124 return str_hash((void *)key);
125 }
126
str_equal_fn(long a,long b,void * ctx)127 static bool str_equal_fn(long a, long b, void *ctx)
128 {
129 return strcmp((void *)a, (void *)b) == 0;
130 }
131
btf_name_of(const struct btf_dump * d,__u32 name_off)132 static const char *btf_name_of(const struct btf_dump *d, __u32 name_off)
133 {
134 return btf__name_by_offset(d->btf, name_off);
135 }
136
btf_dump_printf(const struct btf_dump * d,const char * fmt,...)137 static void btf_dump_printf(const struct btf_dump *d, const char *fmt, ...)
138 {
139 va_list args;
140
141 va_start(args, fmt);
142 d->printf_fn(d->cb_ctx, fmt, args);
143 va_end(args);
144 }
145
146 static int btf_dump_mark_referenced(struct btf_dump *d);
147 static int btf_dump_resize(struct btf_dump *d);
148
btf_dump__new(const struct btf * btf,btf_dump_printf_fn_t printf_fn,void * ctx,const struct btf_dump_opts * opts)149 struct btf_dump *btf_dump__new(const struct btf *btf,
150 btf_dump_printf_fn_t printf_fn,
151 void *ctx,
152 const struct btf_dump_opts *opts)
153 {
154 struct btf_dump *d;
155 int err;
156
157 if (!OPTS_VALID(opts, btf_dump_opts))
158 return libbpf_err_ptr(-EINVAL);
159
160 if (!printf_fn)
161 return libbpf_err_ptr(-EINVAL);
162
163 d = calloc(1, sizeof(struct btf_dump));
164 if (!d)
165 return libbpf_err_ptr(-ENOMEM);
166
167 d->btf = btf;
168 d->printf_fn = printf_fn;
169 d->cb_ctx = ctx;
170 d->ptr_sz = btf__pointer_size(btf) ? : sizeof(void *);
171
172 d->type_names = hashmap__new(str_hash_fn, str_equal_fn, NULL);
173 if (IS_ERR(d->type_names)) {
174 err = PTR_ERR(d->type_names);
175 d->type_names = NULL;
176 goto err;
177 }
178 d->ident_names = hashmap__new(str_hash_fn, str_equal_fn, NULL);
179 if (IS_ERR(d->ident_names)) {
180 err = PTR_ERR(d->ident_names);
181 d->ident_names = NULL;
182 goto err;
183 }
184
185 err = btf_dump_resize(d);
186 if (err)
187 goto err;
188
189 return d;
190 err:
191 btf_dump__free(d);
192 return libbpf_err_ptr(err);
193 }
194
btf_dump_resize(struct btf_dump * d)195 static int btf_dump_resize(struct btf_dump *d)
196 {
197 int err, last_id = btf__type_cnt(d->btf) - 1;
198
199 if (last_id <= d->last_id)
200 return 0;
201
202 if (libbpf_ensure_mem((void **)&d->type_states, &d->type_states_cap,
203 sizeof(*d->type_states), last_id + 1))
204 return -ENOMEM;
205 if (libbpf_ensure_mem((void **)&d->cached_names, &d->cached_names_cap,
206 sizeof(*d->cached_names), last_id + 1))
207 return -ENOMEM;
208
209 if (d->last_id == 0) {
210 /* VOID is special */
211 d->type_states[0].order_state = ORDERED;
212 d->type_states[0].emit_state = EMITTED;
213 }
214
215 /* eagerly determine referenced types for anon enums */
216 err = btf_dump_mark_referenced(d);
217 if (err)
218 return err;
219
220 d->last_id = last_id;
221 return 0;
222 }
223
btf_dump_free_names(struct hashmap * map)224 static void btf_dump_free_names(struct hashmap *map)
225 {
226 size_t bkt;
227 struct hashmap_entry *cur;
228
229 hashmap__for_each_entry(map, cur, bkt)
230 free((void *)cur->pkey);
231
232 hashmap__free(map);
233 }
234
btf_dump__free(struct btf_dump * d)235 void btf_dump__free(struct btf_dump *d)
236 {
237 int i;
238
239 if (IS_ERR_OR_NULL(d))
240 return;
241
242 free(d->type_states);
243 if (d->cached_names) {
244 /* any set cached name is owned by us and should be freed */
245 for (i = 0; i <= d->last_id; i++) {
246 if (d->cached_names[i])
247 free((void *)d->cached_names[i]);
248 }
249 }
250 free(d->cached_names);
251 free(d->emit_queue);
252 free(d->decl_stack);
253 btf_dump_free_names(d->type_names);
254 btf_dump_free_names(d->ident_names);
255
256 free(d);
257 }
258
259 static int btf_dump_order_type(struct btf_dump *d, __u32 id, bool through_ptr);
260 static void btf_dump_emit_type(struct btf_dump *d, __u32 id, __u32 cont_id);
261
262 /*
263 * Dump BTF type in a compilable C syntax, including all the necessary
264 * dependent types, necessary for compilation. If some of the dependent types
265 * were already emitted as part of previous btf_dump__dump_type() invocation
266 * for another type, they won't be emitted again. This API allows callers to
267 * filter out BTF types according to user-defined criterias and emitted only
268 * minimal subset of types, necessary to compile everything. Full struct/union
269 * definitions will still be emitted, even if the only usage is through
270 * pointer and could be satisfied with just a forward declaration.
271 *
272 * Dumping is done in two high-level passes:
273 * 1. Topologically sort type definitions to satisfy C rules of compilation.
274 * 2. Emit type definitions in C syntax.
275 *
276 * Returns 0 on success; <0, otherwise.
277 */
btf_dump__dump_type(struct btf_dump * d,__u32 id)278 int btf_dump__dump_type(struct btf_dump *d, __u32 id)
279 {
280 int err, i;
281
282 if (id >= btf__type_cnt(d->btf))
283 return libbpf_err(-EINVAL);
284
285 err = btf_dump_resize(d);
286 if (err)
287 return libbpf_err(err);
288
289 d->emit_queue_cnt = 0;
290 err = btf_dump_order_type(d, id, false);
291 if (err < 0)
292 return libbpf_err(err);
293
294 for (i = 0; i < d->emit_queue_cnt; i++)
295 btf_dump_emit_type(d, d->emit_queue[i], 0 /*top-level*/);
296
297 return 0;
298 }
299
300 /*
301 * Mark all types that are referenced from any other type. This is used to
302 * determine top-level anonymous enums that need to be emitted as an
303 * independent type declarations.
304 * Anonymous enums come in two flavors: either embedded in a struct's field
305 * definition, in which case they have to be declared inline as part of field
306 * type declaration; or as a top-level anonymous enum, typically used for
307 * declaring global constants. It's impossible to distinguish between two
308 * without knowing whether given enum type was referenced from other type:
309 * top-level anonymous enum won't be referenced by anything, while embedded
310 * one will.
311 */
btf_dump_mark_referenced(struct btf_dump * d)312 static int btf_dump_mark_referenced(struct btf_dump *d)
313 {
314 int i, j, n = btf__type_cnt(d->btf);
315 const struct btf_type *t;
316 __u16 vlen;
317
318 for (i = d->last_id + 1; i < n; i++) {
319 t = btf__type_by_id(d->btf, i);
320 vlen = btf_vlen(t);
321
322 switch (btf_kind(t)) {
323 case BTF_KIND_INT:
324 case BTF_KIND_ENUM:
325 case BTF_KIND_ENUM64:
326 case BTF_KIND_FWD:
327 case BTF_KIND_FLOAT:
328 break;
329
330 case BTF_KIND_VOLATILE:
331 case BTF_KIND_CONST:
332 case BTF_KIND_RESTRICT:
333 case BTF_KIND_PTR:
334 case BTF_KIND_TYPEDEF:
335 case BTF_KIND_FUNC:
336 case BTF_KIND_VAR:
337 case BTF_KIND_DECL_TAG:
338 case BTF_KIND_TYPE_TAG:
339 d->type_states[t->type].referenced = 1;
340 break;
341
342 case BTF_KIND_ARRAY: {
343 const struct btf_array *a = btf_array(t);
344
345 d->type_states[a->index_type].referenced = 1;
346 d->type_states[a->type].referenced = 1;
347 break;
348 }
349 case BTF_KIND_STRUCT:
350 case BTF_KIND_UNION: {
351 const struct btf_member *m = btf_members(t);
352
353 for (j = 0; j < vlen; j++, m++)
354 d->type_states[m->type].referenced = 1;
355 break;
356 }
357 case BTF_KIND_FUNC_PROTO: {
358 const struct btf_param *p = btf_params(t);
359
360 for (j = 0; j < vlen; j++, p++)
361 d->type_states[p->type].referenced = 1;
362 break;
363 }
364 case BTF_KIND_DATASEC: {
365 const struct btf_var_secinfo *v = btf_var_secinfos(t);
366
367 for (j = 0; j < vlen; j++, v++)
368 d->type_states[v->type].referenced = 1;
369 break;
370 }
371 default:
372 return -EINVAL;
373 }
374 }
375 return 0;
376 }
377
btf_dump_add_emit_queue_id(struct btf_dump * d,__u32 id)378 static int btf_dump_add_emit_queue_id(struct btf_dump *d, __u32 id)
379 {
380 __u32 *new_queue;
381 size_t new_cap;
382
383 if (d->emit_queue_cnt >= d->emit_queue_cap) {
384 new_cap = max(16, d->emit_queue_cap * 3 / 2);
385 new_queue = libbpf_reallocarray(d->emit_queue, new_cap, sizeof(new_queue[0]));
386 if (!new_queue)
387 return -ENOMEM;
388 d->emit_queue = new_queue;
389 d->emit_queue_cap = new_cap;
390 }
391
392 d->emit_queue[d->emit_queue_cnt++] = id;
393 return 0;
394 }
395
396 /*
397 * Determine order of emitting dependent types and specified type to satisfy
398 * C compilation rules. This is done through topological sorting with an
399 * additional complication which comes from C rules. The main idea for C is
400 * that if some type is "embedded" into a struct/union, it's size needs to be
401 * known at the time of definition of containing type. E.g., for:
402 *
403 * struct A {};
404 * struct B { struct A x; }
405 *
406 * struct A *HAS* to be defined before struct B, because it's "embedded",
407 * i.e., it is part of struct B layout. But in the following case:
408 *
409 * struct A;
410 * struct B { struct A *x; }
411 * struct A {};
412 *
413 * it's enough to just have a forward declaration of struct A at the time of
414 * struct B definition, as struct B has a pointer to struct A, so the size of
415 * field x is known without knowing struct A size: it's sizeof(void *).
416 *
417 * Unfortunately, there are some trickier cases we need to handle, e.g.:
418 *
419 * struct A {}; // if this was forward-declaration: compilation error
420 * struct B {
421 * struct { // anonymous struct
422 * struct A y;
423 * } *x;
424 * };
425 *
426 * In this case, struct B's field x is a pointer, so it's size is known
427 * regardless of the size of (anonymous) struct it points to. But because this
428 * struct is anonymous and thus defined inline inside struct B, *and* it
429 * embeds struct A, compiler requires full definition of struct A to be known
430 * before struct B can be defined. This creates a transitive dependency
431 * between struct A and struct B. If struct A was forward-declared before
432 * struct B definition and fully defined after struct B definition, that would
433 * trigger compilation error.
434 *
435 * All this means that while we are doing topological sorting on BTF type
436 * graph, we need to determine relationships between different types (graph
437 * nodes):
438 * - weak link (relationship) between X and Y, if Y *CAN* be
439 * forward-declared at the point of X definition;
440 * - strong link, if Y *HAS* to be fully-defined before X can be defined.
441 *
442 * The rule is as follows. Given a chain of BTF types from X to Y, if there is
443 * BTF_KIND_PTR type in the chain and at least one non-anonymous type
444 * Z (excluding X, including Y), then link is weak. Otherwise, it's strong.
445 * Weak/strong relationship is determined recursively during DFS traversal and
446 * is returned as a result from btf_dump_order_type().
447 *
448 * btf_dump_order_type() is trying to avoid unnecessary forward declarations,
449 * but it is not guaranteeing that no extraneous forward declarations will be
450 * emitted.
451 *
452 * To avoid extra work, algorithm marks some of BTF types as ORDERED, when
453 * it's done with them, but not for all (e.g., VOLATILE, CONST, RESTRICT,
454 * ARRAY, FUNC_PROTO), as weak/strong semantics for those depends on the
455 * entire graph path, so depending where from one came to that BTF type, it
456 * might cause weak or strong ordering. For types like STRUCT/UNION/INT/ENUM,
457 * once they are processed, there is no need to do it again, so they are
458 * marked as ORDERED. We can mark PTR as ORDERED as well, as it semi-forces
459 * weak link, unless subsequent referenced STRUCT/UNION/ENUM is anonymous. But
460 * in any case, once those are processed, no need to do it again, as the
461 * result won't change.
462 *
463 * Returns:
464 * - 1, if type is part of strong link (so there is strong topological
465 * ordering requirements);
466 * - 0, if type is part of weak link (so can be satisfied through forward
467 * declaration);
468 * - <0, on error (e.g., unsatisfiable type loop detected).
469 */
btf_dump_order_type(struct btf_dump * d,__u32 id,bool through_ptr)470 static int btf_dump_order_type(struct btf_dump *d, __u32 id, bool through_ptr)
471 {
472 /*
473 * Order state is used to detect strong link cycles, but only for BTF
474 * kinds that are or could be an independent definition (i.e.,
475 * stand-alone fwd decl, enum, typedef, struct, union). Ptrs, arrays,
476 * func_protos, modifiers are just means to get to these definitions.
477 * Int/void don't need definitions, they are assumed to be always
478 * properly defined. We also ignore datasec, var, and funcs for now.
479 * So for all non-defining kinds, we never even set ordering state,
480 * for defining kinds we set ORDERING and subsequently ORDERED if it
481 * forms a strong link.
482 */
483 struct btf_dump_type_aux_state *tstate = &d->type_states[id];
484 const struct btf_type *t;
485 __u16 vlen;
486 int err, i;
487
488 /* return true, letting typedefs know that it's ok to be emitted */
489 if (tstate->order_state == ORDERED)
490 return 1;
491
492 t = btf__type_by_id(d->btf, id);
493
494 if (tstate->order_state == ORDERING) {
495 /* type loop, but resolvable through fwd declaration */
496 if (btf_is_composite(t) && through_ptr && t->name_off != 0)
497 return 0;
498 pr_warn("unsatisfiable type cycle, id:[%u]\n", id);
499 return -ELOOP;
500 }
501
502 switch (btf_kind(t)) {
503 case BTF_KIND_INT:
504 case BTF_KIND_FLOAT:
505 tstate->order_state = ORDERED;
506 return 0;
507
508 case BTF_KIND_PTR:
509 err = btf_dump_order_type(d, t->type, true);
510 tstate->order_state = ORDERED;
511 return err;
512
513 case BTF_KIND_ARRAY:
514 return btf_dump_order_type(d, btf_array(t)->type, false);
515
516 case BTF_KIND_STRUCT:
517 case BTF_KIND_UNION: {
518 const struct btf_member *m = btf_members(t);
519 /*
520 * struct/union is part of strong link, only if it's embedded
521 * (so no ptr in a path) or it's anonymous (so has to be
522 * defined inline, even if declared through ptr)
523 */
524 if (through_ptr && t->name_off != 0)
525 return 0;
526
527 tstate->order_state = ORDERING;
528
529 vlen = btf_vlen(t);
530 for (i = 0; i < vlen; i++, m++) {
531 err = btf_dump_order_type(d, m->type, false);
532 if (err < 0)
533 return err;
534 }
535
536 if (t->name_off != 0) {
537 err = btf_dump_add_emit_queue_id(d, id);
538 if (err < 0)
539 return err;
540 }
541
542 tstate->order_state = ORDERED;
543 return 1;
544 }
545 case BTF_KIND_ENUM:
546 case BTF_KIND_ENUM64:
547 case BTF_KIND_FWD:
548 /*
549 * non-anonymous or non-referenced enums are top-level
550 * declarations and should be emitted. Same logic can be
551 * applied to FWDs, it won't hurt anyways.
552 */
553 if (t->name_off != 0 || !tstate->referenced) {
554 err = btf_dump_add_emit_queue_id(d, id);
555 if (err)
556 return err;
557 }
558 tstate->order_state = ORDERED;
559 return 1;
560
561 case BTF_KIND_TYPEDEF: {
562 int is_strong;
563
564 is_strong = btf_dump_order_type(d, t->type, through_ptr);
565 if (is_strong < 0)
566 return is_strong;
567
568 /* typedef is similar to struct/union w.r.t. fwd-decls */
569 if (through_ptr && !is_strong)
570 return 0;
571
572 /* typedef is always a named definition */
573 err = btf_dump_add_emit_queue_id(d, id);
574 if (err)
575 return err;
576
577 d->type_states[id].order_state = ORDERED;
578 return 1;
579 }
580 case BTF_KIND_VOLATILE:
581 case BTF_KIND_CONST:
582 case BTF_KIND_RESTRICT:
583 case BTF_KIND_TYPE_TAG:
584 return btf_dump_order_type(d, t->type, through_ptr);
585
586 case BTF_KIND_FUNC_PROTO: {
587 const struct btf_param *p = btf_params(t);
588 bool is_strong;
589
590 err = btf_dump_order_type(d, t->type, through_ptr);
591 if (err < 0)
592 return err;
593 is_strong = err > 0;
594
595 vlen = btf_vlen(t);
596 for (i = 0; i < vlen; i++, p++) {
597 err = btf_dump_order_type(d, p->type, through_ptr);
598 if (err < 0)
599 return err;
600 if (err > 0)
601 is_strong = true;
602 }
603 return is_strong;
604 }
605 case BTF_KIND_FUNC:
606 case BTF_KIND_VAR:
607 case BTF_KIND_DATASEC:
608 case BTF_KIND_DECL_TAG:
609 d->type_states[id].order_state = ORDERED;
610 return 0;
611
612 default:
613 return -EINVAL;
614 }
615 }
616
617 static void btf_dump_emit_missing_aliases(struct btf_dump *d, __u32 id,
618 const struct btf_type *t);
619
620 static void btf_dump_emit_struct_fwd(struct btf_dump *d, __u32 id,
621 const struct btf_type *t);
622 static void btf_dump_emit_struct_def(struct btf_dump *d, __u32 id,
623 const struct btf_type *t, int lvl);
624
625 static void btf_dump_emit_enum_fwd(struct btf_dump *d, __u32 id,
626 const struct btf_type *t);
627 static void btf_dump_emit_enum_def(struct btf_dump *d, __u32 id,
628 const struct btf_type *t, int lvl);
629
630 static void btf_dump_emit_fwd_def(struct btf_dump *d, __u32 id,
631 const struct btf_type *t);
632
633 static void btf_dump_emit_typedef_def(struct btf_dump *d, __u32 id,
634 const struct btf_type *t, int lvl);
635
636 /* a local view into a shared stack */
637 struct id_stack {
638 const __u32 *ids;
639 int cnt;
640 };
641
642 static void btf_dump_emit_type_decl(struct btf_dump *d, __u32 id,
643 const char *fname, int lvl);
644 static void btf_dump_emit_type_chain(struct btf_dump *d,
645 struct id_stack *decl_stack,
646 const char *fname, int lvl);
647
648 static const char *btf_dump_type_name(struct btf_dump *d, __u32 id);
649 static const char *btf_dump_ident_name(struct btf_dump *d, __u32 id);
650 static size_t btf_dump_name_dups(struct btf_dump *d, struct hashmap *name_map,
651 const char *orig_name);
652
btf_dump_is_blacklisted(struct btf_dump * d,__u32 id)653 static bool btf_dump_is_blacklisted(struct btf_dump *d, __u32 id)
654 {
655 const struct btf_type *t = btf__type_by_id(d->btf, id);
656
657 /* __builtin_va_list is a compiler built-in, which causes compilation
658 * errors, when compiling w/ different compiler, then used to compile
659 * original code (e.g., GCC to compile kernel, Clang to use generated
660 * C header from BTF). As it is built-in, it should be already defined
661 * properly internally in compiler.
662 */
663 if (t->name_off == 0)
664 return false;
665 return strcmp(btf_name_of(d, t->name_off), "__builtin_va_list") == 0;
666 }
667
668 /*
669 * Emit C-syntax definitions of types from chains of BTF types.
670 *
671 * High-level handling of determining necessary forward declarations are handled
672 * by btf_dump_emit_type() itself, but all nitty-gritty details of emitting type
673 * declarations/definitions in C syntax are handled by a combo of
674 * btf_dump_emit_type_decl()/btf_dump_emit_type_chain() w/ delegation to
675 * corresponding btf_dump_emit_*_{def,fwd}() functions.
676 *
677 * We also keep track of "containing struct/union type ID" to determine when
678 * we reference it from inside and thus can avoid emitting unnecessary forward
679 * declaration.
680 *
681 * This algorithm is designed in such a way, that even if some error occurs
682 * (either technical, e.g., out of memory, or logical, i.e., malformed BTF
683 * that doesn't comply to C rules completely), algorithm will try to proceed
684 * and produce as much meaningful output as possible.
685 */
btf_dump_emit_type(struct btf_dump * d,__u32 id,__u32 cont_id)686 static void btf_dump_emit_type(struct btf_dump *d, __u32 id, __u32 cont_id)
687 {
688 struct btf_dump_type_aux_state *tstate = &d->type_states[id];
689 bool top_level_def = cont_id == 0;
690 const struct btf_type *t;
691 __u16 kind;
692
693 if (tstate->emit_state == EMITTED)
694 return;
695
696 t = btf__type_by_id(d->btf, id);
697 kind = btf_kind(t);
698
699 if (tstate->emit_state == EMITTING) {
700 if (tstate->fwd_emitted)
701 return;
702
703 switch (kind) {
704 case BTF_KIND_STRUCT:
705 case BTF_KIND_UNION:
706 /*
707 * if we are referencing a struct/union that we are
708 * part of - then no need for fwd declaration
709 */
710 if (id == cont_id)
711 return;
712 if (t->name_off == 0) {
713 pr_warn("anonymous struct/union loop, id:[%u]\n",
714 id);
715 return;
716 }
717 btf_dump_emit_struct_fwd(d, id, t);
718 btf_dump_printf(d, ";\n\n");
719 tstate->fwd_emitted = 1;
720 break;
721 case BTF_KIND_TYPEDEF:
722 /*
723 * for typedef fwd_emitted means typedef definition
724 * was emitted, but it can be used only for "weak"
725 * references through pointer only, not for embedding
726 */
727 if (!btf_dump_is_blacklisted(d, id)) {
728 btf_dump_emit_typedef_def(d, id, t, 0);
729 btf_dump_printf(d, ";\n\n");
730 }
731 tstate->fwd_emitted = 1;
732 break;
733 default:
734 break;
735 }
736
737 return;
738 }
739
740 switch (kind) {
741 case BTF_KIND_INT:
742 /* Emit type alias definitions if necessary */
743 btf_dump_emit_missing_aliases(d, id, t);
744
745 tstate->emit_state = EMITTED;
746 break;
747 case BTF_KIND_ENUM:
748 case BTF_KIND_ENUM64:
749 if (top_level_def) {
750 btf_dump_emit_enum_def(d, id, t, 0);
751 btf_dump_printf(d, ";\n\n");
752 }
753 tstate->emit_state = EMITTED;
754 break;
755 case BTF_KIND_PTR:
756 case BTF_KIND_VOLATILE:
757 case BTF_KIND_CONST:
758 case BTF_KIND_RESTRICT:
759 case BTF_KIND_TYPE_TAG:
760 btf_dump_emit_type(d, t->type, cont_id);
761 break;
762 case BTF_KIND_ARRAY:
763 btf_dump_emit_type(d, btf_array(t)->type, cont_id);
764 break;
765 case BTF_KIND_FWD:
766 btf_dump_emit_fwd_def(d, id, t);
767 btf_dump_printf(d, ";\n\n");
768 tstate->emit_state = EMITTED;
769 break;
770 case BTF_KIND_TYPEDEF:
771 tstate->emit_state = EMITTING;
772 btf_dump_emit_type(d, t->type, id);
773 /*
774 * typedef can server as both definition and forward
775 * declaration; at this stage someone depends on
776 * typedef as a forward declaration (refers to it
777 * through pointer), so unless we already did it,
778 * emit typedef as a forward declaration
779 */
780 if (!tstate->fwd_emitted && !btf_dump_is_blacklisted(d, id)) {
781 btf_dump_emit_typedef_def(d, id, t, 0);
782 btf_dump_printf(d, ";\n\n");
783 }
784 tstate->emit_state = EMITTED;
785 break;
786 case BTF_KIND_STRUCT:
787 case BTF_KIND_UNION:
788 tstate->emit_state = EMITTING;
789 /* if it's a top-level struct/union definition or struct/union
790 * is anonymous, then in C we'll be emitting all fields and
791 * their types (as opposed to just `struct X`), so we need to
792 * make sure that all types, referenced from struct/union
793 * members have necessary forward-declarations, where
794 * applicable
795 */
796 if (top_level_def || t->name_off == 0) {
797 const struct btf_member *m = btf_members(t);
798 __u16 vlen = btf_vlen(t);
799 int i, new_cont_id;
800
801 new_cont_id = t->name_off == 0 ? cont_id : id;
802 for (i = 0; i < vlen; i++, m++)
803 btf_dump_emit_type(d, m->type, new_cont_id);
804 } else if (!tstate->fwd_emitted && id != cont_id) {
805 btf_dump_emit_struct_fwd(d, id, t);
806 btf_dump_printf(d, ";\n\n");
807 tstate->fwd_emitted = 1;
808 }
809
810 if (top_level_def) {
811 btf_dump_emit_struct_def(d, id, t, 0);
812 btf_dump_printf(d, ";\n\n");
813 tstate->emit_state = EMITTED;
814 } else {
815 tstate->emit_state = NOT_EMITTED;
816 }
817 break;
818 case BTF_KIND_FUNC_PROTO: {
819 const struct btf_param *p = btf_params(t);
820 __u16 n = btf_vlen(t);
821 int i;
822
823 btf_dump_emit_type(d, t->type, cont_id);
824 for (i = 0; i < n; i++, p++)
825 btf_dump_emit_type(d, p->type, cont_id);
826
827 break;
828 }
829 default:
830 break;
831 }
832 }
833
btf_is_struct_packed(const struct btf * btf,__u32 id,const struct btf_type * t)834 static bool btf_is_struct_packed(const struct btf *btf, __u32 id,
835 const struct btf_type *t)
836 {
837 const struct btf_member *m;
838 int max_align = 1, align, i, bit_sz;
839 __u16 vlen;
840
841 m = btf_members(t);
842 vlen = btf_vlen(t);
843 /* all non-bitfield fields have to be naturally aligned */
844 for (i = 0; i < vlen; i++, m++) {
845 align = btf__align_of(btf, m->type);
846 bit_sz = btf_member_bitfield_size(t, i);
847 if (align && bit_sz == 0 && m->offset % (8 * align) != 0)
848 return true;
849 max_align = max(align, max_align);
850 }
851 /* size of a non-packed struct has to be a multiple of its alignment */
852 if (t->size % max_align != 0)
853 return true;
854 /*
855 * if original struct was marked as packed, but its layout is
856 * naturally aligned, we'll detect that it's not packed
857 */
858 return false;
859 }
860
btf_dump_emit_bit_padding(const struct btf_dump * d,int cur_off,int next_off,int next_align,bool in_bitfield,int lvl)861 static void btf_dump_emit_bit_padding(const struct btf_dump *d,
862 int cur_off, int next_off, int next_align,
863 bool in_bitfield, int lvl)
864 {
865 const struct {
866 const char *name;
867 int bits;
868 } pads[] = {
869 {"long", d->ptr_sz * 8}, {"int", 32}, {"short", 16}, {"char", 8}
870 };
871 int new_off = 0, pad_bits = 0, bits, i;
872 const char *pad_type = NULL;
873
874 if (cur_off >= next_off)
875 return; /* no gap */
876
877 /* For filling out padding we want to take advantage of
878 * natural alignment rules to minimize unnecessary explicit
879 * padding. First, we find the largest type (among long, int,
880 * short, or char) that can be used to force naturally aligned
881 * boundary. Once determined, we'll use such type to fill in
882 * the remaining padding gap. In some cases we can rely on
883 * compiler filling some gaps, but sometimes we need to force
884 * alignment to close natural alignment with markers like
885 * `long: 0` (this is always the case for bitfields). Note
886 * that even if struct itself has, let's say 4-byte alignment
887 * (i.e., it only uses up to int-aligned types), using `long:
888 * X;` explicit padding doesn't actually change struct's
889 * overall alignment requirements, but compiler does take into
890 * account that type's (long, in this example) natural
891 * alignment requirements when adding implicit padding. We use
892 * this fact heavily and don't worry about ruining correct
893 * struct alignment requirement.
894 */
895 for (i = 0; i < ARRAY_SIZE(pads); i++) {
896 pad_bits = pads[i].bits;
897 pad_type = pads[i].name;
898
899 new_off = roundup(cur_off, pad_bits);
900 if (new_off <= next_off)
901 break;
902 }
903
904 if (new_off > cur_off && new_off <= next_off) {
905 /* We need explicit `<type>: 0` aligning mark if next
906 * field is right on alignment offset and its
907 * alignment requirement is less strict than <type>'s
908 * alignment (so compiler won't naturally align to the
909 * offset we expect), or if subsequent `<type>: X`,
910 * will actually completely fit in the remaining hole,
911 * making compiler basically ignore `<type>: X`
912 * completely.
913 */
914 if (in_bitfield ||
915 (new_off == next_off && roundup(cur_off, next_align * 8) != new_off) ||
916 (new_off != next_off && next_off - new_off <= new_off - cur_off))
917 /* but for bitfields we'll emit explicit bit count */
918 btf_dump_printf(d, "\n%s%s: %d;", pfx(lvl), pad_type,
919 in_bitfield ? new_off - cur_off : 0);
920 cur_off = new_off;
921 }
922
923 /* Now we know we start at naturally aligned offset for a chosen
924 * padding type (long, int, short, or char), and so the rest is just
925 * a straightforward filling of remaining padding gap with full
926 * `<type>: sizeof(<type>);` markers, except for the last one, which
927 * might need smaller than sizeof(<type>) padding.
928 */
929 while (cur_off != next_off) {
930 bits = min(next_off - cur_off, pad_bits);
931 if (bits == pad_bits) {
932 btf_dump_printf(d, "\n%s%s: %d;", pfx(lvl), pad_type, pad_bits);
933 cur_off += bits;
934 continue;
935 }
936 /* For the remainder padding that doesn't cover entire
937 * pad_type bit length, we pick the smallest necessary type.
938 * This is pure aesthetics, we could have just used `long`,
939 * but having smallest necessary one communicates better the
940 * scale of the padding gap.
941 */
942 for (i = ARRAY_SIZE(pads) - 1; i >= 0; i--) {
943 pad_type = pads[i].name;
944 pad_bits = pads[i].bits;
945 if (pad_bits < bits)
946 continue;
947
948 btf_dump_printf(d, "\n%s%s: %d;", pfx(lvl), pad_type, bits);
949 cur_off += bits;
950 break;
951 }
952 }
953 }
954
btf_dump_emit_struct_fwd(struct btf_dump * d,__u32 id,const struct btf_type * t)955 static void btf_dump_emit_struct_fwd(struct btf_dump *d, __u32 id,
956 const struct btf_type *t)
957 {
958 btf_dump_printf(d, "%s%s%s",
959 btf_is_struct(t) ? "struct" : "union",
960 t->name_off ? " " : "",
961 btf_dump_type_name(d, id));
962 }
963
btf_dump_emit_struct_def(struct btf_dump * d,__u32 id,const struct btf_type * t,int lvl)964 static void btf_dump_emit_struct_def(struct btf_dump *d,
965 __u32 id,
966 const struct btf_type *t,
967 int lvl)
968 {
969 const struct btf_member *m = btf_members(t);
970 bool is_struct = btf_is_struct(t);
971 bool packed, prev_bitfield = false;
972 int align, i, off = 0;
973 __u16 vlen = btf_vlen(t);
974
975 align = btf__align_of(d->btf, id);
976 packed = is_struct ? btf_is_struct_packed(d->btf, id, t) : 0;
977
978 btf_dump_printf(d, "%s%s%s {",
979 is_struct ? "struct" : "union",
980 t->name_off ? " " : "",
981 btf_dump_type_name(d, id));
982
983 for (i = 0; i < vlen; i++, m++) {
984 const char *fname;
985 int m_off, m_sz, m_align;
986 bool in_bitfield;
987
988 fname = btf_name_of(d, m->name_off);
989 m_sz = btf_member_bitfield_size(t, i);
990 m_off = btf_member_bit_offset(t, i);
991 m_align = packed ? 1 : btf__align_of(d->btf, m->type);
992
993 in_bitfield = prev_bitfield && m_sz != 0;
994
995 btf_dump_emit_bit_padding(d, off, m_off, m_align, in_bitfield, lvl + 1);
996 btf_dump_printf(d, "\n%s", pfx(lvl + 1));
997 btf_dump_emit_type_decl(d, m->type, fname, lvl + 1);
998
999 if (m_sz) {
1000 btf_dump_printf(d, ": %d", m_sz);
1001 off = m_off + m_sz;
1002 prev_bitfield = true;
1003 } else {
1004 m_sz = max((__s64)0, btf__resolve_size(d->btf, m->type));
1005 off = m_off + m_sz * 8;
1006 prev_bitfield = false;
1007 }
1008
1009 btf_dump_printf(d, ";");
1010 }
1011
1012 /* pad at the end, if necessary */
1013 if (is_struct)
1014 btf_dump_emit_bit_padding(d, off, t->size * 8, align, false, lvl + 1);
1015
1016 /*
1017 * Keep `struct empty {}` on a single line,
1018 * only print newline when there are regular or padding fields.
1019 */
1020 if (vlen || t->size) {
1021 btf_dump_printf(d, "\n");
1022 btf_dump_printf(d, "%s}", pfx(lvl));
1023 } else {
1024 btf_dump_printf(d, "}");
1025 }
1026 if (packed)
1027 btf_dump_printf(d, " __attribute__((packed))");
1028 }
1029
1030 static const char *missing_base_types[][2] = {
1031 /*
1032 * GCC emits typedefs to its internal __PolyX_t types when compiling Arm
1033 * SIMD intrinsics. Alias them to standard base types.
1034 */
1035 { "__Poly8_t", "unsigned char" },
1036 { "__Poly16_t", "unsigned short" },
1037 { "__Poly64_t", "unsigned long long" },
1038 { "__Poly128_t", "unsigned __int128" },
1039 };
1040
btf_dump_emit_missing_aliases(struct btf_dump * d,__u32 id,const struct btf_type * t)1041 static void btf_dump_emit_missing_aliases(struct btf_dump *d, __u32 id,
1042 const struct btf_type *t)
1043 {
1044 const char *name = btf_dump_type_name(d, id);
1045 int i;
1046
1047 for (i = 0; i < ARRAY_SIZE(missing_base_types); i++) {
1048 if (strcmp(name, missing_base_types[i][0]) == 0) {
1049 btf_dump_printf(d, "typedef %s %s;\n\n",
1050 missing_base_types[i][1], name);
1051 break;
1052 }
1053 }
1054 }
1055
btf_dump_emit_enum_fwd(struct btf_dump * d,__u32 id,const struct btf_type * t)1056 static void btf_dump_emit_enum_fwd(struct btf_dump *d, __u32 id,
1057 const struct btf_type *t)
1058 {
1059 btf_dump_printf(d, "enum %s", btf_dump_type_name(d, id));
1060 }
1061
btf_dump_emit_enum32_val(struct btf_dump * d,const struct btf_type * t,int lvl,__u16 vlen)1062 static void btf_dump_emit_enum32_val(struct btf_dump *d,
1063 const struct btf_type *t,
1064 int lvl, __u16 vlen)
1065 {
1066 const struct btf_enum *v = btf_enum(t);
1067 bool is_signed = btf_kflag(t);
1068 const char *fmt_str;
1069 const char *name;
1070 size_t dup_cnt;
1071 int i;
1072
1073 for (i = 0; i < vlen; i++, v++) {
1074 name = btf_name_of(d, v->name_off);
1075 /* enumerators share namespace with typedef idents */
1076 dup_cnt = btf_dump_name_dups(d, d->ident_names, name);
1077 if (dup_cnt > 1) {
1078 fmt_str = is_signed ? "\n%s%s___%zd = %d," : "\n%s%s___%zd = %u,";
1079 btf_dump_printf(d, fmt_str, pfx(lvl + 1), name, dup_cnt, v->val);
1080 } else {
1081 fmt_str = is_signed ? "\n%s%s = %d," : "\n%s%s = %u,";
1082 btf_dump_printf(d, fmt_str, pfx(lvl + 1), name, v->val);
1083 }
1084 }
1085 }
1086
btf_dump_emit_enum64_val(struct btf_dump * d,const struct btf_type * t,int lvl,__u16 vlen)1087 static void btf_dump_emit_enum64_val(struct btf_dump *d,
1088 const struct btf_type *t,
1089 int lvl, __u16 vlen)
1090 {
1091 const struct btf_enum64 *v = btf_enum64(t);
1092 bool is_signed = btf_kflag(t);
1093 const char *fmt_str;
1094 const char *name;
1095 size_t dup_cnt;
1096 __u64 val;
1097 int i;
1098
1099 for (i = 0; i < vlen; i++, v++) {
1100 name = btf_name_of(d, v->name_off);
1101 dup_cnt = btf_dump_name_dups(d, d->ident_names, name);
1102 val = btf_enum64_value(v);
1103 if (dup_cnt > 1) {
1104 fmt_str = is_signed ? "\n%s%s___%zd = %lldLL,"
1105 : "\n%s%s___%zd = %lluULL,";
1106 btf_dump_printf(d, fmt_str,
1107 pfx(lvl + 1), name, dup_cnt,
1108 (unsigned long long)val);
1109 } else {
1110 fmt_str = is_signed ? "\n%s%s = %lldLL,"
1111 : "\n%s%s = %lluULL,";
1112 btf_dump_printf(d, fmt_str,
1113 pfx(lvl + 1), name,
1114 (unsigned long long)val);
1115 }
1116 }
1117 }
btf_dump_emit_enum_def(struct btf_dump * d,__u32 id,const struct btf_type * t,int lvl)1118 static void btf_dump_emit_enum_def(struct btf_dump *d, __u32 id,
1119 const struct btf_type *t,
1120 int lvl)
1121 {
1122 __u16 vlen = btf_vlen(t);
1123
1124 btf_dump_printf(d, "enum%s%s",
1125 t->name_off ? " " : "",
1126 btf_dump_type_name(d, id));
1127
1128 if (!vlen)
1129 return;
1130
1131 btf_dump_printf(d, " {");
1132 if (btf_is_enum(t))
1133 btf_dump_emit_enum32_val(d, t, lvl, vlen);
1134 else
1135 btf_dump_emit_enum64_val(d, t, lvl, vlen);
1136 btf_dump_printf(d, "\n%s}", pfx(lvl));
1137
1138 /* special case enums with special sizes */
1139 if (t->size == 1) {
1140 /* one-byte enums can be forced with mode(byte) attribute */
1141 btf_dump_printf(d, " __attribute__((mode(byte)))");
1142 } else if (t->size == 8 && d->ptr_sz == 8) {
1143 /* enum can be 8-byte sized if one of the enumerator values
1144 * doesn't fit in 32-bit integer, or by adding mode(word)
1145 * attribute (but probably only on 64-bit architectures); do
1146 * our best here to try to satisfy the contract without adding
1147 * unnecessary attributes
1148 */
1149 bool needs_word_mode;
1150
1151 if (btf_is_enum(t)) {
1152 /* enum can't represent 64-bit values, so we need word mode */
1153 needs_word_mode = true;
1154 } else {
1155 /* enum64 needs mode(word) if none of its values has
1156 * non-zero upper 32-bits (which means that all values
1157 * fit in 32-bit integers and won't cause compiler to
1158 * bump enum to be 64-bit naturally
1159 */
1160 int i;
1161
1162 needs_word_mode = true;
1163 for (i = 0; i < vlen; i++) {
1164 if (btf_enum64(t)[i].val_hi32 != 0) {
1165 needs_word_mode = false;
1166 break;
1167 }
1168 }
1169 }
1170 if (needs_word_mode)
1171 btf_dump_printf(d, " __attribute__((mode(word)))");
1172 }
1173
1174 }
1175
btf_dump_emit_fwd_def(struct btf_dump * d,__u32 id,const struct btf_type * t)1176 static void btf_dump_emit_fwd_def(struct btf_dump *d, __u32 id,
1177 const struct btf_type *t)
1178 {
1179 const char *name = btf_dump_type_name(d, id);
1180
1181 if (btf_kflag(t))
1182 btf_dump_printf(d, "union %s", name);
1183 else
1184 btf_dump_printf(d, "struct %s", name);
1185 }
1186
btf_dump_emit_typedef_def(struct btf_dump * d,__u32 id,const struct btf_type * t,int lvl)1187 static void btf_dump_emit_typedef_def(struct btf_dump *d, __u32 id,
1188 const struct btf_type *t, int lvl)
1189 {
1190 const char *name = btf_dump_ident_name(d, id);
1191
1192 /*
1193 * Old GCC versions are emitting invalid typedef for __gnuc_va_list
1194 * pointing to VOID. This generates warnings from btf_dump() and
1195 * results in uncompilable header file, so we are fixing it up here
1196 * with valid typedef into __builtin_va_list.
1197 */
1198 if (t->type == 0 && strcmp(name, "__gnuc_va_list") == 0) {
1199 btf_dump_printf(d, "typedef __builtin_va_list __gnuc_va_list");
1200 return;
1201 }
1202
1203 btf_dump_printf(d, "typedef ");
1204 btf_dump_emit_type_decl(d, t->type, name, lvl);
1205 }
1206
btf_dump_push_decl_stack_id(struct btf_dump * d,__u32 id)1207 static int btf_dump_push_decl_stack_id(struct btf_dump *d, __u32 id)
1208 {
1209 __u32 *new_stack;
1210 size_t new_cap;
1211
1212 if (d->decl_stack_cnt >= d->decl_stack_cap) {
1213 new_cap = max(16, d->decl_stack_cap * 3 / 2);
1214 new_stack = libbpf_reallocarray(d->decl_stack, new_cap, sizeof(new_stack[0]));
1215 if (!new_stack)
1216 return -ENOMEM;
1217 d->decl_stack = new_stack;
1218 d->decl_stack_cap = new_cap;
1219 }
1220
1221 d->decl_stack[d->decl_stack_cnt++] = id;
1222
1223 return 0;
1224 }
1225
1226 /*
1227 * Emit type declaration (e.g., field type declaration in a struct or argument
1228 * declaration in function prototype) in correct C syntax.
1229 *
1230 * For most types it's trivial, but there are few quirky type declaration
1231 * cases worth mentioning:
1232 * - function prototypes (especially nesting of function prototypes);
1233 * - arrays;
1234 * - const/volatile/restrict for pointers vs other types.
1235 *
1236 * For a good discussion of *PARSING* C syntax (as a human), see
1237 * Peter van der Linden's "Expert C Programming: Deep C Secrets",
1238 * Ch.3 "Unscrambling Declarations in C".
1239 *
1240 * It won't help with BTF to C conversion much, though, as it's an opposite
1241 * problem. So we came up with this algorithm in reverse to van der Linden's
1242 * parsing algorithm. It goes from structured BTF representation of type
1243 * declaration to a valid compilable C syntax.
1244 *
1245 * For instance, consider this C typedef:
1246 * typedef const int * const * arr[10] arr_t;
1247 * It will be represented in BTF with this chain of BTF types:
1248 * [typedef] -> [array] -> [ptr] -> [const] -> [ptr] -> [const] -> [int]
1249 *
1250 * Notice how [const] modifier always goes before type it modifies in BTF type
1251 * graph, but in C syntax, const/volatile/restrict modifiers are written to
1252 * the right of pointers, but to the left of other types. There are also other
1253 * quirks, like function pointers, arrays of them, functions returning other
1254 * functions, etc.
1255 *
1256 * We handle that by pushing all the types to a stack, until we hit "terminal"
1257 * type (int/enum/struct/union/fwd). Then depending on the kind of a type on
1258 * top of a stack, modifiers are handled differently. Array/function pointers
1259 * have also wildly different syntax and how nesting of them are done. See
1260 * code for authoritative definition.
1261 *
1262 * To avoid allocating new stack for each independent chain of BTF types, we
1263 * share one bigger stack, with each chain working only on its own local view
1264 * of a stack frame. Some care is required to "pop" stack frames after
1265 * processing type declaration chain.
1266 */
btf_dump__emit_type_decl(struct btf_dump * d,__u32 id,const struct btf_dump_emit_type_decl_opts * opts)1267 int btf_dump__emit_type_decl(struct btf_dump *d, __u32 id,
1268 const struct btf_dump_emit_type_decl_opts *opts)
1269 {
1270 const char *fname;
1271 int lvl, err;
1272
1273 if (!OPTS_VALID(opts, btf_dump_emit_type_decl_opts))
1274 return libbpf_err(-EINVAL);
1275
1276 err = btf_dump_resize(d);
1277 if (err)
1278 return libbpf_err(err);
1279
1280 fname = OPTS_GET(opts, field_name, "");
1281 lvl = OPTS_GET(opts, indent_level, 0);
1282 d->strip_mods = OPTS_GET(opts, strip_mods, false);
1283 btf_dump_emit_type_decl(d, id, fname, lvl);
1284 d->strip_mods = false;
1285 return 0;
1286 }
1287
btf_dump_emit_type_decl(struct btf_dump * d,__u32 id,const char * fname,int lvl)1288 static void btf_dump_emit_type_decl(struct btf_dump *d, __u32 id,
1289 const char *fname, int lvl)
1290 {
1291 struct id_stack decl_stack;
1292 const struct btf_type *t;
1293 int err, stack_start;
1294
1295 stack_start = d->decl_stack_cnt;
1296 for (;;) {
1297 t = btf__type_by_id(d->btf, id);
1298 if (d->strip_mods && btf_is_mod(t))
1299 goto skip_mod;
1300
1301 err = btf_dump_push_decl_stack_id(d, id);
1302 if (err < 0) {
1303 /*
1304 * if we don't have enough memory for entire type decl
1305 * chain, restore stack, emit warning, and try to
1306 * proceed nevertheless
1307 */
1308 pr_warn("not enough memory for decl stack: %s\n", errstr(err));
1309 d->decl_stack_cnt = stack_start;
1310 return;
1311 }
1312 skip_mod:
1313 /* VOID */
1314 if (id == 0)
1315 break;
1316
1317 switch (btf_kind(t)) {
1318 case BTF_KIND_PTR:
1319 case BTF_KIND_VOLATILE:
1320 case BTF_KIND_CONST:
1321 case BTF_KIND_RESTRICT:
1322 case BTF_KIND_FUNC_PROTO:
1323 case BTF_KIND_TYPE_TAG:
1324 id = t->type;
1325 break;
1326 case BTF_KIND_ARRAY:
1327 id = btf_array(t)->type;
1328 break;
1329 case BTF_KIND_INT:
1330 case BTF_KIND_ENUM:
1331 case BTF_KIND_ENUM64:
1332 case BTF_KIND_FWD:
1333 case BTF_KIND_STRUCT:
1334 case BTF_KIND_UNION:
1335 case BTF_KIND_TYPEDEF:
1336 case BTF_KIND_FLOAT:
1337 goto done;
1338 default:
1339 pr_warn("unexpected type in decl chain, kind:%u, id:[%u]\n",
1340 btf_kind(t), id);
1341 goto done;
1342 }
1343 }
1344 done:
1345 /*
1346 * We might be inside a chain of declarations (e.g., array of function
1347 * pointers returning anonymous (so inlined) structs, having another
1348 * array field). Each of those needs its own "stack frame" to handle
1349 * emitting of declarations. Those stack frames are non-overlapping
1350 * portions of shared btf_dump->decl_stack. To make it a bit nicer to
1351 * handle this set of nested stacks, we create a view corresponding to
1352 * our own "stack frame" and work with it as an independent stack.
1353 * We'll need to clean up after emit_type_chain() returns, though.
1354 */
1355 decl_stack.ids = d->decl_stack + stack_start;
1356 decl_stack.cnt = d->decl_stack_cnt - stack_start;
1357 btf_dump_emit_type_chain(d, &decl_stack, fname, lvl);
1358 /*
1359 * emit_type_chain() guarantees that it will pop its entire decl_stack
1360 * frame before returning. But it works with a read-only view into
1361 * decl_stack, so it doesn't actually pop anything from the
1362 * perspective of shared btf_dump->decl_stack, per se. We need to
1363 * reset decl_stack state to how it was before us to avoid it growing
1364 * all the time.
1365 */
1366 d->decl_stack_cnt = stack_start;
1367 }
1368
btf_dump_emit_mods(struct btf_dump * d,struct id_stack * decl_stack)1369 static void btf_dump_emit_mods(struct btf_dump *d, struct id_stack *decl_stack)
1370 {
1371 const struct btf_type *t;
1372 __u32 id;
1373
1374 while (decl_stack->cnt) {
1375 id = decl_stack->ids[decl_stack->cnt - 1];
1376 t = btf__type_by_id(d->btf, id);
1377
1378 switch (btf_kind(t)) {
1379 case BTF_KIND_VOLATILE:
1380 btf_dump_printf(d, "volatile ");
1381 break;
1382 case BTF_KIND_CONST:
1383 btf_dump_printf(d, "const ");
1384 break;
1385 case BTF_KIND_RESTRICT:
1386 btf_dump_printf(d, "restrict ");
1387 break;
1388 default:
1389 return;
1390 }
1391 decl_stack->cnt--;
1392 }
1393 }
1394
btf_dump_drop_mods(struct btf_dump * d,struct id_stack * decl_stack)1395 static void btf_dump_drop_mods(struct btf_dump *d, struct id_stack *decl_stack)
1396 {
1397 const struct btf_type *t;
1398 __u32 id;
1399
1400 while (decl_stack->cnt) {
1401 id = decl_stack->ids[decl_stack->cnt - 1];
1402 t = btf__type_by_id(d->btf, id);
1403 if (!btf_is_mod(t))
1404 return;
1405 decl_stack->cnt--;
1406 }
1407 }
1408
btf_dump_emit_name(const struct btf_dump * d,const char * name,bool last_was_ptr)1409 static void btf_dump_emit_name(const struct btf_dump *d,
1410 const char *name, bool last_was_ptr)
1411 {
1412 bool separate = name[0] && !last_was_ptr;
1413
1414 btf_dump_printf(d, "%s%s", separate ? " " : "", name);
1415 }
1416
btf_dump_emit_type_chain(struct btf_dump * d,struct id_stack * decls,const char * fname,int lvl)1417 static void btf_dump_emit_type_chain(struct btf_dump *d,
1418 struct id_stack *decls,
1419 const char *fname, int lvl)
1420 {
1421 /*
1422 * last_was_ptr is used to determine if we need to separate pointer
1423 * asterisk (*) from previous part of type signature with space, so
1424 * that we get `int ***`, instead of `int * * *`. We default to true
1425 * for cases where we have single pointer in a chain. E.g., in ptr ->
1426 * func_proto case. func_proto will start a new emit_type_chain call
1427 * with just ptr, which should be emitted as (*) or (*<fname>), so we
1428 * don't want to prepend space for that last pointer.
1429 */
1430 bool last_was_ptr = true;
1431 const struct btf_type *t;
1432 const char *name;
1433 __u16 kind;
1434 __u32 id;
1435
1436 while (decls->cnt) {
1437 id = decls->ids[--decls->cnt];
1438 if (id == 0) {
1439 /* VOID is a special snowflake */
1440 btf_dump_emit_mods(d, decls);
1441 btf_dump_printf(d, "void");
1442 last_was_ptr = false;
1443 continue;
1444 }
1445
1446 t = btf__type_by_id(d->btf, id);
1447 kind = btf_kind(t);
1448
1449 switch (kind) {
1450 case BTF_KIND_INT:
1451 case BTF_KIND_FLOAT:
1452 btf_dump_emit_mods(d, decls);
1453 name = btf_name_of(d, t->name_off);
1454 btf_dump_printf(d, "%s", name);
1455 break;
1456 case BTF_KIND_STRUCT:
1457 case BTF_KIND_UNION:
1458 btf_dump_emit_mods(d, decls);
1459 /* inline anonymous struct/union */
1460 if (t->name_off == 0 && !d->skip_anon_defs)
1461 btf_dump_emit_struct_def(d, id, t, lvl);
1462 else
1463 btf_dump_emit_struct_fwd(d, id, t);
1464 break;
1465 case BTF_KIND_ENUM:
1466 case BTF_KIND_ENUM64:
1467 btf_dump_emit_mods(d, decls);
1468 /* inline anonymous enum */
1469 if (t->name_off == 0 && !d->skip_anon_defs)
1470 btf_dump_emit_enum_def(d, id, t, lvl);
1471 else
1472 btf_dump_emit_enum_fwd(d, id, t);
1473 break;
1474 case BTF_KIND_FWD:
1475 btf_dump_emit_mods(d, decls);
1476 btf_dump_emit_fwd_def(d, id, t);
1477 break;
1478 case BTF_KIND_TYPEDEF:
1479 btf_dump_emit_mods(d, decls);
1480 btf_dump_printf(d, "%s", btf_dump_ident_name(d, id));
1481 break;
1482 case BTF_KIND_PTR:
1483 btf_dump_printf(d, "%s", last_was_ptr ? "*" : " *");
1484 break;
1485 case BTF_KIND_VOLATILE:
1486 btf_dump_printf(d, " volatile");
1487 break;
1488 case BTF_KIND_CONST:
1489 btf_dump_printf(d, " const");
1490 break;
1491 case BTF_KIND_RESTRICT:
1492 btf_dump_printf(d, " restrict");
1493 break;
1494 case BTF_KIND_TYPE_TAG:
1495 btf_dump_emit_mods(d, decls);
1496 name = btf_name_of(d, t->name_off);
1497 if (btf_kflag(t))
1498 btf_dump_printf(d, " __attribute__((%s))", name);
1499 else
1500 btf_dump_printf(d, " __attribute__((btf_type_tag(\"%s\")))", name);
1501 break;
1502 case BTF_KIND_ARRAY: {
1503 const struct btf_array *a = btf_array(t);
1504 const struct btf_type *next_t;
1505 __u32 next_id;
1506 bool multidim;
1507 /*
1508 * GCC has a bug
1509 * (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=8354)
1510 * which causes it to emit extra const/volatile
1511 * modifiers for an array, if array's element type has
1512 * const/volatile modifiers. Clang doesn't do that.
1513 * In general, it doesn't seem very meaningful to have
1514 * a const/volatile modifier for array, so we are
1515 * going to silently skip them here.
1516 */
1517 btf_dump_drop_mods(d, decls);
1518
1519 if (decls->cnt == 0) {
1520 btf_dump_emit_name(d, fname, last_was_ptr);
1521 btf_dump_printf(d, "[%u]", a->nelems);
1522 return;
1523 }
1524
1525 next_id = decls->ids[decls->cnt - 1];
1526 next_t = btf__type_by_id(d->btf, next_id);
1527 multidim = btf_is_array(next_t);
1528 /* we need space if we have named non-pointer */
1529 if (fname[0] && !last_was_ptr)
1530 btf_dump_printf(d, " ");
1531 /* no parentheses for multi-dimensional array */
1532 if (!multidim)
1533 btf_dump_printf(d, "(");
1534 btf_dump_emit_type_chain(d, decls, fname, lvl);
1535 if (!multidim)
1536 btf_dump_printf(d, ")");
1537 btf_dump_printf(d, "[%u]", a->nelems);
1538 return;
1539 }
1540 case BTF_KIND_FUNC_PROTO: {
1541 const struct btf_param *p = btf_params(t);
1542 __u16 vlen = btf_vlen(t);
1543 int i;
1544
1545 /*
1546 * GCC emits extra volatile qualifier for
1547 * __attribute__((noreturn)) function pointers. Clang
1548 * doesn't do it. It's a GCC quirk for backwards
1549 * compatibility with code written for GCC <2.5. So,
1550 * similarly to extra qualifiers for array, just drop
1551 * them, instead of handling them.
1552 */
1553 btf_dump_drop_mods(d, decls);
1554 if (decls->cnt) {
1555 btf_dump_printf(d, " (");
1556 btf_dump_emit_type_chain(d, decls, fname, lvl);
1557 btf_dump_printf(d, ")");
1558 } else {
1559 btf_dump_emit_name(d, fname, last_was_ptr);
1560 }
1561 btf_dump_printf(d, "(");
1562 /*
1563 * Clang for BPF target generates func_proto with no
1564 * args as a func_proto with a single void arg (e.g.,
1565 * `int (*f)(void)` vs just `int (*f)()`). We are
1566 * going to emit valid empty args (void) syntax for
1567 * such case. Similarly and conveniently, valid
1568 * no args case can be special-cased here as well.
1569 */
1570 if (vlen == 0 || (vlen == 1 && p->type == 0)) {
1571 btf_dump_printf(d, "void)");
1572 return;
1573 }
1574
1575 for (i = 0; i < vlen; i++, p++) {
1576 if (i > 0)
1577 btf_dump_printf(d, ", ");
1578
1579 /* last arg of type void is vararg */
1580 if (i == vlen - 1 && p->type == 0) {
1581 btf_dump_printf(d, "...");
1582 break;
1583 }
1584
1585 name = btf_name_of(d, p->name_off);
1586 btf_dump_emit_type_decl(d, p->type, name, lvl);
1587 }
1588
1589 btf_dump_printf(d, ")");
1590 return;
1591 }
1592 default:
1593 pr_warn("unexpected type in decl chain, kind:%u, id:[%u]\n",
1594 kind, id);
1595 return;
1596 }
1597
1598 last_was_ptr = kind == BTF_KIND_PTR;
1599 }
1600
1601 btf_dump_emit_name(d, fname, last_was_ptr);
1602 }
1603
1604 /* show type name as (type_name) */
btf_dump_emit_type_cast(struct btf_dump * d,__u32 id,bool top_level)1605 static void btf_dump_emit_type_cast(struct btf_dump *d, __u32 id,
1606 bool top_level)
1607 {
1608 const struct btf_type *t;
1609
1610 /* for array members, we don't bother emitting type name for each
1611 * member to avoid the redundancy of
1612 * .name = (char[4])[(char)'f',(char)'o',(char)'o',]
1613 */
1614 if (d->typed_dump->is_array_member)
1615 return;
1616
1617 /* avoid type name specification for variable/section; it will be done
1618 * for the associated variable value(s).
1619 */
1620 t = btf__type_by_id(d->btf, id);
1621 if (btf_is_var(t) || btf_is_datasec(t))
1622 return;
1623
1624 if (top_level)
1625 btf_dump_printf(d, "(");
1626
1627 d->skip_anon_defs = true;
1628 d->strip_mods = true;
1629 btf_dump_emit_type_decl(d, id, "", 0);
1630 d->strip_mods = false;
1631 d->skip_anon_defs = false;
1632
1633 if (top_level)
1634 btf_dump_printf(d, ")");
1635 }
1636
1637 /* 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)1638 static size_t btf_dump_name_dups(struct btf_dump *d, struct hashmap *name_map,
1639 const char *orig_name)
1640 {
1641 char *old_name, *new_name;
1642 size_t dup_cnt = 0;
1643 int err;
1644
1645 new_name = strdup(orig_name);
1646 if (!new_name)
1647 return 1;
1648
1649 (void)hashmap__find(name_map, orig_name, &dup_cnt);
1650 dup_cnt++;
1651
1652 err = hashmap__set(name_map, new_name, dup_cnt, &old_name, NULL);
1653 if (err)
1654 free(new_name);
1655
1656 free(old_name);
1657
1658 return dup_cnt;
1659 }
1660
btf_dump_resolve_name(struct btf_dump * d,__u32 id,struct hashmap * name_map)1661 static const char *btf_dump_resolve_name(struct btf_dump *d, __u32 id,
1662 struct hashmap *name_map)
1663 {
1664 struct btf_dump_type_aux_state *s = &d->type_states[id];
1665 const struct btf_type *t = btf__type_by_id(d->btf, id);
1666 const char *orig_name = btf_name_of(d, t->name_off);
1667 const char **cached_name = &d->cached_names[id];
1668 size_t dup_cnt;
1669
1670 if (t->name_off == 0)
1671 return "";
1672
1673 if (s->name_resolved)
1674 return *cached_name ? *cached_name : orig_name;
1675
1676 if (btf_is_fwd(t) || (btf_is_enum(t) && btf_vlen(t) == 0)) {
1677 s->name_resolved = 1;
1678 return orig_name;
1679 }
1680
1681 dup_cnt = btf_dump_name_dups(d, name_map, orig_name);
1682 if (dup_cnt > 1) {
1683 const size_t max_len = 256;
1684 char new_name[max_len];
1685
1686 snprintf(new_name, max_len, "%s___%zu", orig_name, dup_cnt);
1687 *cached_name = strdup(new_name);
1688 }
1689
1690 s->name_resolved = 1;
1691 return *cached_name ? *cached_name : orig_name;
1692 }
1693
btf_dump_type_name(struct btf_dump * d,__u32 id)1694 static const char *btf_dump_type_name(struct btf_dump *d, __u32 id)
1695 {
1696 return btf_dump_resolve_name(d, id, d->type_names);
1697 }
1698
btf_dump_ident_name(struct btf_dump * d,__u32 id)1699 static const char *btf_dump_ident_name(struct btf_dump *d, __u32 id)
1700 {
1701 return btf_dump_resolve_name(d, id, d->ident_names);
1702 }
1703
1704 static int btf_dump_dump_type_data(struct btf_dump *d,
1705 const char *fname,
1706 const struct btf_type *t,
1707 __u32 id,
1708 const void *data,
1709 __u8 bits_offset,
1710 __u8 bit_sz);
1711
btf_dump_data_newline(struct btf_dump * d)1712 static const char *btf_dump_data_newline(struct btf_dump *d)
1713 {
1714 return d->typed_dump->compact || d->typed_dump->depth == 0 ? "" : "\n";
1715 }
1716
btf_dump_data_delim(struct btf_dump * d)1717 static const char *btf_dump_data_delim(struct btf_dump *d)
1718 {
1719 return d->typed_dump->depth == 0 ? "" : ",";
1720 }
1721
btf_dump_data_pfx(struct btf_dump * d)1722 static void btf_dump_data_pfx(struct btf_dump *d)
1723 {
1724 int i, lvl = d->typed_dump->indent_lvl + d->typed_dump->depth;
1725
1726 if (d->typed_dump->compact)
1727 return;
1728
1729 for (i = 0; i < lvl; i++)
1730 btf_dump_printf(d, "%s", d->typed_dump->indent_str);
1731 }
1732
1733 /* A macro is used here as btf_type_value[s]() appends format specifiers
1734 * to the format specifier passed in; these do the work of appending
1735 * delimiters etc while the caller simply has to specify the type values
1736 * in the format specifier + value(s).
1737 */
1738 #define btf_dump_type_values(d, fmt, ...) \
1739 btf_dump_printf(d, fmt "%s%s", \
1740 ##__VA_ARGS__, \
1741 btf_dump_data_delim(d), \
1742 btf_dump_data_newline(d))
1743
btf_dump_unsupported_data(struct btf_dump * d,const struct btf_type * t,__u32 id)1744 static int btf_dump_unsupported_data(struct btf_dump *d,
1745 const struct btf_type *t,
1746 __u32 id)
1747 {
1748 btf_dump_printf(d, "<unsupported kind:%u>", btf_kind(t));
1749 return -ENOTSUP;
1750 }
1751
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)1752 static int btf_dump_get_bitfield_value(struct btf_dump *d,
1753 const struct btf_type *t,
1754 const void *data,
1755 __u8 bits_offset,
1756 __u8 bit_sz,
1757 __u64 *value)
1758 {
1759 __u16 left_shift_bits, right_shift_bits;
1760 const __u8 *bytes = data;
1761 __u8 nr_copy_bits;
1762 __u64 num = 0;
1763 int i;
1764
1765 /* Maximum supported bitfield size is 64 bits */
1766 if (t->size > 8) {
1767 pr_warn("unexpected bitfield size %d\n", t->size);
1768 return -EINVAL;
1769 }
1770
1771 /* Bitfield value retrieval is done in two steps; first relevant bytes are
1772 * stored in num, then we left/right shift num to eliminate irrelevant bits.
1773 */
1774 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
1775 for (i = t->size - 1; i >= 0; i--)
1776 num = num * 256 + bytes[i];
1777 nr_copy_bits = bit_sz + bits_offset;
1778 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
1779 for (i = 0; i < t->size; i++)
1780 num = num * 256 + bytes[i];
1781 nr_copy_bits = t->size * 8 - bits_offset;
1782 #else
1783 # error "Unrecognized __BYTE_ORDER__"
1784 #endif
1785 left_shift_bits = 64 - nr_copy_bits;
1786 right_shift_bits = 64 - bit_sz;
1787
1788 *value = (num << left_shift_bits) >> right_shift_bits;
1789
1790 return 0;
1791 }
1792
btf_dump_bitfield_check_zero(struct btf_dump * d,const struct btf_type * t,const void * data,__u8 bits_offset,__u8 bit_sz)1793 static int btf_dump_bitfield_check_zero(struct btf_dump *d,
1794 const struct btf_type *t,
1795 const void *data,
1796 __u8 bits_offset,
1797 __u8 bit_sz)
1798 {
1799 __u64 check_num;
1800 int err;
1801
1802 err = btf_dump_get_bitfield_value(d, t, data, bits_offset, bit_sz, &check_num);
1803 if (err)
1804 return err;
1805 if (check_num == 0)
1806 return -ENODATA;
1807 return 0;
1808 }
1809
btf_dump_bitfield_data(struct btf_dump * d,const struct btf_type * t,const void * data,__u8 bits_offset,__u8 bit_sz)1810 static int btf_dump_bitfield_data(struct btf_dump *d,
1811 const struct btf_type *t,
1812 const void *data,
1813 __u8 bits_offset,
1814 __u8 bit_sz)
1815 {
1816 __u64 print_num;
1817 int err;
1818
1819 err = btf_dump_get_bitfield_value(d, t, data, bits_offset, bit_sz, &print_num);
1820 if (err)
1821 return err;
1822
1823 btf_dump_type_values(d, "0x%llx", (unsigned long long)print_num);
1824
1825 return 0;
1826 }
1827
1828 /* ints, floats and ptrs */
btf_dump_base_type_check_zero(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data)1829 static int btf_dump_base_type_check_zero(struct btf_dump *d,
1830 const struct btf_type *t,
1831 __u32 id,
1832 const void *data)
1833 {
1834 static __u8 bytecmp[16] = {};
1835 int nr_bytes;
1836
1837 /* For pointer types, pointer size is not defined on a per-type basis.
1838 * On dump creation however, we store the pointer size.
1839 */
1840 if (btf_kind(t) == BTF_KIND_PTR)
1841 nr_bytes = d->ptr_sz;
1842 else
1843 nr_bytes = t->size;
1844
1845 if (nr_bytes < 1 || nr_bytes > 16) {
1846 pr_warn("unexpected size %d for id [%u]\n", nr_bytes, id);
1847 return -EINVAL;
1848 }
1849
1850 if (memcmp(data, bytecmp, nr_bytes) == 0)
1851 return -ENODATA;
1852 return 0;
1853 }
1854
ptr_is_aligned(const struct btf * btf,__u32 type_id,const void * data)1855 static bool ptr_is_aligned(const struct btf *btf, __u32 type_id,
1856 const void *data)
1857 {
1858 int alignment = btf__align_of(btf, type_id);
1859
1860 if (alignment == 0)
1861 return false;
1862
1863 return ((uintptr_t)data) % alignment == 0;
1864 }
1865
btf_dump_int_data(struct btf_dump * d,const struct btf_type * t,__u32 type_id,const void * data,__u8 bits_offset)1866 static int btf_dump_int_data(struct btf_dump *d,
1867 const struct btf_type *t,
1868 __u32 type_id,
1869 const void *data,
1870 __u8 bits_offset)
1871 {
1872 __u8 encoding = btf_int_encoding(t);
1873 bool sign = encoding & BTF_INT_SIGNED;
1874 char buf[16] __attribute__((aligned(16)));
1875 int sz = t->size;
1876
1877 if (sz == 0 || sz > sizeof(buf)) {
1878 pr_warn("unexpected size %d for id [%u]\n", sz, type_id);
1879 return -EINVAL;
1880 }
1881
1882 /* handle packed int data - accesses of integers not aligned on
1883 * int boundaries can cause problems on some platforms.
1884 */
1885 if (!ptr_is_aligned(d->btf, type_id, data)) {
1886 memcpy(buf, data, sz);
1887 data = buf;
1888 }
1889
1890 switch (sz) {
1891 case 16: {
1892 const __u64 *ints = data;
1893 __u64 lsi, msi;
1894
1895 /* avoid use of __int128 as some 32-bit platforms do not
1896 * support it.
1897 */
1898 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
1899 lsi = ints[0];
1900 msi = ints[1];
1901 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
1902 lsi = ints[1];
1903 msi = ints[0];
1904 #else
1905 # error "Unrecognized __BYTE_ORDER__"
1906 #endif
1907 if (msi == 0)
1908 btf_dump_type_values(d, "0x%llx", (unsigned long long)lsi);
1909 else
1910 btf_dump_type_values(d, "0x%llx%016llx", (unsigned long long)msi,
1911 (unsigned long long)lsi);
1912 break;
1913 }
1914 case 8:
1915 if (sign)
1916 btf_dump_type_values(d, "%lld", *(long long *)data);
1917 else
1918 btf_dump_type_values(d, "%llu", *(unsigned long long *)data);
1919 break;
1920 case 4:
1921 if (sign)
1922 btf_dump_type_values(d, "%d", *(__s32 *)data);
1923 else
1924 btf_dump_type_values(d, "%u", *(__u32 *)data);
1925 break;
1926 case 2:
1927 if (sign)
1928 btf_dump_type_values(d, "%d", *(__s16 *)data);
1929 else
1930 btf_dump_type_values(d, "%u", *(__u16 *)data);
1931 break;
1932 case 1:
1933 if (d->typed_dump->is_array_char) {
1934 /* check for null terminator */
1935 if (d->typed_dump->is_array_terminated)
1936 break;
1937 if (*(char *)data == '\0') {
1938 btf_dump_type_values(d, "'\\0'");
1939 d->typed_dump->is_array_terminated = true;
1940 break;
1941 }
1942 if (isprint(*(char *)data)) {
1943 btf_dump_type_values(d, "'%c'", *(char *)data);
1944 break;
1945 }
1946 }
1947 if (sign)
1948 btf_dump_type_values(d, "%d", *(__s8 *)data);
1949 else
1950 btf_dump_type_values(d, "%u", *(__u8 *)data);
1951 break;
1952 default:
1953 pr_warn("unexpected sz %d for id [%u]\n", sz, type_id);
1954 return -EINVAL;
1955 }
1956 return 0;
1957 }
1958
1959 union float_data {
1960 long double ld;
1961 double d;
1962 float f;
1963 };
1964
btf_dump_float_data(struct btf_dump * d,const struct btf_type * t,__u32 type_id,const void * data)1965 static int btf_dump_float_data(struct btf_dump *d,
1966 const struct btf_type *t,
1967 __u32 type_id,
1968 const void *data)
1969 {
1970 const union float_data *flp = data;
1971 union float_data fl;
1972 int sz = t->size;
1973
1974 /* handle unaligned data; copy to local union */
1975 if (!ptr_is_aligned(d->btf, type_id, data)) {
1976 memcpy(&fl, data, sz);
1977 flp = &fl;
1978 }
1979
1980 switch (sz) {
1981 case 16:
1982 btf_dump_type_values(d, "%Lf", flp->ld);
1983 break;
1984 case 8:
1985 btf_dump_type_values(d, "%lf", flp->d);
1986 break;
1987 case 4:
1988 btf_dump_type_values(d, "%f", flp->f);
1989 break;
1990 default:
1991 pr_warn("unexpected size %d for id [%u]\n", sz, type_id);
1992 return -EINVAL;
1993 }
1994 return 0;
1995 }
1996
btf_dump_var_data(struct btf_dump * d,const struct btf_type * v,__u32 id,const void * data)1997 static int btf_dump_var_data(struct btf_dump *d,
1998 const struct btf_type *v,
1999 __u32 id,
2000 const void *data)
2001 {
2002 enum btf_func_linkage linkage = btf_var(v)->linkage;
2003 const struct btf_type *t;
2004 const char *l;
2005 __u32 type_id;
2006
2007 switch (linkage) {
2008 case BTF_FUNC_STATIC:
2009 l = "static ";
2010 break;
2011 case BTF_FUNC_EXTERN:
2012 l = "extern ";
2013 break;
2014 case BTF_FUNC_GLOBAL:
2015 default:
2016 l = "";
2017 break;
2018 }
2019
2020 /* format of output here is [linkage] [type] [varname] = (type)value,
2021 * for example "static int cpu_profile_flip = (int)1"
2022 */
2023 btf_dump_printf(d, "%s", l);
2024 type_id = v->type;
2025 t = btf__type_by_id(d->btf, type_id);
2026 btf_dump_emit_type_cast(d, type_id, false);
2027 btf_dump_printf(d, " %s = ", btf_name_of(d, v->name_off));
2028 return btf_dump_dump_type_data(d, NULL, t, type_id, data, 0, 0);
2029 }
2030
btf_dump_array_data(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data)2031 static int btf_dump_array_data(struct btf_dump *d,
2032 const struct btf_type *t,
2033 __u32 id,
2034 const void *data)
2035 {
2036 const struct btf_array *array = btf_array(t);
2037 const struct btf_type *elem_type;
2038 __u32 i, elem_type_id;
2039 __s64 elem_size;
2040 bool is_array_member;
2041 bool is_array_terminated;
2042
2043 elem_type_id = array->type;
2044 elem_type = skip_mods_and_typedefs(d->btf, elem_type_id, NULL);
2045 elem_size = btf__resolve_size(d->btf, elem_type_id);
2046 if (elem_size <= 0) {
2047 pr_warn("unexpected elem size %zd for array type [%u]\n",
2048 (ssize_t)elem_size, id);
2049 return -EINVAL;
2050 }
2051
2052 if (btf_is_int(elem_type)) {
2053 /*
2054 * BTF_INT_CHAR encoding never seems to be set for
2055 * char arrays, so if size is 1 and element is
2056 * printable as a char, we'll do that.
2057 */
2058 if (elem_size == 1)
2059 d->typed_dump->is_array_char = true;
2060 }
2061
2062 /* note that we increment depth before calling btf_dump_print() below;
2063 * this is intentional. btf_dump_data_newline() will not print a
2064 * newline for depth 0 (since this leaves us with trailing newlines
2065 * at the end of typed display), so depth is incremented first.
2066 * For similar reasons, we decrement depth before showing the closing
2067 * parenthesis.
2068 */
2069 d->typed_dump->depth++;
2070 btf_dump_printf(d, "[%s", btf_dump_data_newline(d));
2071
2072 /* may be a multidimensional array, so store current "is array member"
2073 * status so we can restore it correctly later.
2074 */
2075 is_array_member = d->typed_dump->is_array_member;
2076 d->typed_dump->is_array_member = true;
2077 is_array_terminated = d->typed_dump->is_array_terminated;
2078 d->typed_dump->is_array_terminated = false;
2079 for (i = 0; i < array->nelems; i++, data += elem_size) {
2080 if (d->typed_dump->is_array_terminated)
2081 break;
2082 btf_dump_dump_type_data(d, NULL, elem_type, elem_type_id, data, 0, 0);
2083 }
2084 d->typed_dump->is_array_member = is_array_member;
2085 d->typed_dump->is_array_terminated = is_array_terminated;
2086 d->typed_dump->depth--;
2087 btf_dump_data_pfx(d);
2088 btf_dump_type_values(d, "]");
2089
2090 return 0;
2091 }
2092
btf_dump_struct_data(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data)2093 static int btf_dump_struct_data(struct btf_dump *d,
2094 const struct btf_type *t,
2095 __u32 id,
2096 const void *data)
2097 {
2098 const struct btf_member *m = btf_members(t);
2099 __u16 n = btf_vlen(t);
2100 int i, err = 0;
2101
2102 /* note that we increment depth before calling btf_dump_print() below;
2103 * this is intentional. btf_dump_data_newline() will not print a
2104 * newline for depth 0 (since this leaves us with trailing newlines
2105 * at the end of typed display), so depth is incremented first.
2106 * For similar reasons, we decrement depth before showing the closing
2107 * parenthesis.
2108 */
2109 d->typed_dump->depth++;
2110 btf_dump_printf(d, "{%s", btf_dump_data_newline(d));
2111
2112 for (i = 0; i < n; i++, m++) {
2113 const struct btf_type *mtype;
2114 const char *mname;
2115 __u32 moffset;
2116 __u8 bit_sz;
2117
2118 mtype = btf__type_by_id(d->btf, m->type);
2119 mname = btf_name_of(d, m->name_off);
2120 moffset = btf_member_bit_offset(t, i);
2121
2122 bit_sz = btf_member_bitfield_size(t, i);
2123 err = btf_dump_dump_type_data(d, mname, mtype, m->type, data + moffset / 8,
2124 moffset % 8, bit_sz);
2125 if (err < 0)
2126 return err;
2127 }
2128 d->typed_dump->depth--;
2129 btf_dump_data_pfx(d);
2130 btf_dump_type_values(d, "}");
2131 return err;
2132 }
2133
2134 union ptr_data {
2135 unsigned int p;
2136 unsigned long long lp;
2137 };
2138
btf_dump_ptr_data(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data)2139 static int btf_dump_ptr_data(struct btf_dump *d,
2140 const struct btf_type *t,
2141 __u32 id,
2142 const void *data)
2143 {
2144 if (ptr_is_aligned(d->btf, id, data) && d->ptr_sz == sizeof(void *)) {
2145 btf_dump_type_values(d, "%p", *(void **)data);
2146 } else {
2147 union ptr_data pt;
2148
2149 memcpy(&pt, data, d->ptr_sz);
2150 if (d->ptr_sz == 4)
2151 btf_dump_type_values(d, "0x%x", pt.p);
2152 else
2153 btf_dump_type_values(d, "0x%llx", pt.lp);
2154 }
2155 return 0;
2156 }
2157
btf_dump_get_enum_value(struct btf_dump * d,const struct btf_type * t,const void * data,__u32 id,__s64 * value)2158 static int btf_dump_get_enum_value(struct btf_dump *d,
2159 const struct btf_type *t,
2160 const void *data,
2161 __u32 id,
2162 __s64 *value)
2163 {
2164 bool is_signed = btf_kflag(t);
2165
2166 if (!ptr_is_aligned(d->btf, id, data)) {
2167 __u64 val;
2168 int err;
2169
2170 err = btf_dump_get_bitfield_value(d, t, data, 0, 0, &val);
2171 if (err)
2172 return err;
2173 *value = (__s64)val;
2174 return 0;
2175 }
2176
2177 switch (t->size) {
2178 case 8:
2179 *value = *(__s64 *)data;
2180 return 0;
2181 case 4:
2182 *value = is_signed ? (__s64)*(__s32 *)data : *(__u32 *)data;
2183 return 0;
2184 case 2:
2185 *value = is_signed ? *(__s16 *)data : *(__u16 *)data;
2186 return 0;
2187 case 1:
2188 *value = is_signed ? *(__s8 *)data : *(__u8 *)data;
2189 return 0;
2190 default:
2191 pr_warn("unexpected size %d for enum, id:[%u]\n", t->size, id);
2192 return -EINVAL;
2193 }
2194 }
2195
btf_dump_enum_data(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data)2196 static int btf_dump_enum_data(struct btf_dump *d,
2197 const struct btf_type *t,
2198 __u32 id,
2199 const void *data)
2200 {
2201 bool is_signed;
2202 __s64 value;
2203 int i, err;
2204
2205 err = btf_dump_get_enum_value(d, t, data, id, &value);
2206 if (err)
2207 return err;
2208
2209 is_signed = btf_kflag(t);
2210 if (btf_is_enum(t)) {
2211 const struct btf_enum *e;
2212
2213 for (i = 0, e = btf_enum(t); i < btf_vlen(t); i++, e++) {
2214 if (value != e->val)
2215 continue;
2216 btf_dump_type_values(d, "%s", btf_name_of(d, e->name_off));
2217 return 0;
2218 }
2219
2220 btf_dump_type_values(d, is_signed ? "%d" : "%u", value);
2221 } else {
2222 const struct btf_enum64 *e;
2223
2224 for (i = 0, e = btf_enum64(t); i < btf_vlen(t); i++, e++) {
2225 if (value != btf_enum64_value(e))
2226 continue;
2227 btf_dump_type_values(d, "%s", btf_name_of(d, e->name_off));
2228 return 0;
2229 }
2230
2231 btf_dump_type_values(d, is_signed ? "%lldLL" : "%lluULL",
2232 (unsigned long long)value);
2233 }
2234 return 0;
2235 }
2236
btf_dump_datasec_data(struct btf_dump * d,const struct btf_type * t,__u32 id,const void * data)2237 static int btf_dump_datasec_data(struct btf_dump *d,
2238 const struct btf_type *t,
2239 __u32 id,
2240 const void *data)
2241 {
2242 const struct btf_var_secinfo *vsi;
2243 const struct btf_type *var;
2244 __u32 i;
2245 int err;
2246
2247 btf_dump_type_values(d, "SEC(\"%s\") ", btf_name_of(d, t->name_off));
2248
2249 for (i = 0, vsi = btf_var_secinfos(t); i < btf_vlen(t); i++, vsi++) {
2250 var = btf__type_by_id(d->btf, vsi->type);
2251 err = btf_dump_dump_type_data(d, NULL, var, vsi->type, data + vsi->offset, 0, 0);
2252 if (err < 0)
2253 return err;
2254 btf_dump_printf(d, ";");
2255 }
2256 return 0;
2257 }
2258
2259 /* 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)2260 static int btf_dump_type_data_check_overflow(struct btf_dump *d,
2261 const struct btf_type *t,
2262 __u32 id,
2263 const void *data,
2264 __u8 bits_offset,
2265 __u8 bit_sz)
2266 {
2267 __s64 size;
2268
2269 if (bit_sz) {
2270 /* bits_offset is at most 7. bit_sz is at most 128. */
2271 __u8 nr_bytes = (bits_offset + bit_sz + 7) / 8;
2272
2273 /* When bit_sz is non zero, it is called from
2274 * btf_dump_struct_data() where it only cares about
2275 * negative error value.
2276 * Return nr_bytes in success case to make it
2277 * consistent as the regular integer case below.
2278 */
2279 return data + nr_bytes > d->typed_dump->data_end ? -E2BIG : nr_bytes;
2280 }
2281
2282 size = btf__resolve_size(d->btf, id);
2283
2284 if (size < 0 || size >= INT_MAX) {
2285 pr_warn("unexpected size [%zu] for id [%u]\n",
2286 (size_t)size, id);
2287 return -EINVAL;
2288 }
2289
2290 /* Only do overflow checking for base types; we do not want to
2291 * avoid showing part of a struct, union or array, even if we
2292 * do not have enough data to show the full object. By
2293 * restricting overflow checking to base types we can ensure
2294 * that partial display succeeds, while avoiding overflowing
2295 * and using bogus data for display.
2296 */
2297 t = skip_mods_and_typedefs(d->btf, id, NULL);
2298 if (!t) {
2299 pr_warn("unexpected error skipping mods/typedefs for id [%u]\n",
2300 id);
2301 return -EINVAL;
2302 }
2303
2304 switch (btf_kind(t)) {
2305 case BTF_KIND_INT:
2306 case BTF_KIND_FLOAT:
2307 case BTF_KIND_PTR:
2308 case BTF_KIND_ENUM:
2309 case BTF_KIND_ENUM64:
2310 if (data + bits_offset / 8 + size > d->typed_dump->data_end)
2311 return -E2BIG;
2312 break;
2313 default:
2314 break;
2315 }
2316 return (int)size;
2317 }
2318
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)2319 static int btf_dump_type_data_check_zero(struct btf_dump *d,
2320 const struct btf_type *t,
2321 __u32 id,
2322 const void *data,
2323 __u8 bits_offset,
2324 __u8 bit_sz)
2325 {
2326 __s64 value;
2327 int i, err;
2328
2329 /* toplevel exceptions; we show zero values if
2330 * - we ask for them (emit_zeros)
2331 * - if we are at top-level so we see "struct empty { }"
2332 * - or if we are an array member and the array is non-empty and
2333 * not a char array; we don't want to be in a situation where we
2334 * have an integer array 0, 1, 0, 1 and only show non-zero values.
2335 * If the array contains zeroes only, or is a char array starting
2336 * with a '\0', the array-level check_zero() will prevent showing it;
2337 * we are concerned with determining zero value at the array member
2338 * level here.
2339 */
2340 if (d->typed_dump->emit_zeroes || d->typed_dump->depth == 0 ||
2341 (d->typed_dump->is_array_member &&
2342 !d->typed_dump->is_array_char))
2343 return 0;
2344
2345 t = skip_mods_and_typedefs(d->btf, id, NULL);
2346
2347 switch (btf_kind(t)) {
2348 case BTF_KIND_INT:
2349 if (bit_sz)
2350 return btf_dump_bitfield_check_zero(d, t, data, bits_offset, bit_sz);
2351 return btf_dump_base_type_check_zero(d, t, id, data);
2352 case BTF_KIND_FLOAT:
2353 case BTF_KIND_PTR:
2354 return btf_dump_base_type_check_zero(d, t, id, data);
2355 case BTF_KIND_ARRAY: {
2356 const struct btf_array *array = btf_array(t);
2357 const struct btf_type *elem_type;
2358 __u32 elem_type_id, elem_size;
2359 bool ischar;
2360
2361 elem_type_id = array->type;
2362 elem_size = btf__resolve_size(d->btf, elem_type_id);
2363 elem_type = skip_mods_and_typedefs(d->btf, elem_type_id, NULL);
2364
2365 ischar = btf_is_int(elem_type) && elem_size == 1;
2366
2367 /* check all elements; if _any_ element is nonzero, all
2368 * of array is displayed. We make an exception however
2369 * for char arrays where the first element is 0; these
2370 * are considered zeroed also, even if later elements are
2371 * non-zero because the string is terminated.
2372 */
2373 for (i = 0; i < array->nelems; i++) {
2374 if (i == 0 && ischar && *(char *)data == 0)
2375 return -ENODATA;
2376 err = btf_dump_type_data_check_zero(d, elem_type,
2377 elem_type_id,
2378 data +
2379 (i * elem_size),
2380 bits_offset, 0);
2381 if (err != -ENODATA)
2382 return err;
2383 }
2384 return -ENODATA;
2385 }
2386 case BTF_KIND_STRUCT:
2387 case BTF_KIND_UNION: {
2388 const struct btf_member *m = btf_members(t);
2389 __u16 n = btf_vlen(t);
2390
2391 /* if any struct/union member is non-zero, the struct/union
2392 * is considered non-zero and dumped.
2393 */
2394 for (i = 0; i < n; i++, m++) {
2395 const struct btf_type *mtype;
2396 __u32 moffset;
2397
2398 mtype = btf__type_by_id(d->btf, m->type);
2399 moffset = btf_member_bit_offset(t, i);
2400
2401 /* btf_int_bits() does not store member bitfield size;
2402 * bitfield size needs to be stored here so int display
2403 * of member can retrieve it.
2404 */
2405 bit_sz = btf_member_bitfield_size(t, i);
2406 err = btf_dump_type_data_check_zero(d, mtype, m->type, data + moffset / 8,
2407 moffset % 8, bit_sz);
2408 if (err != ENODATA)
2409 return err;
2410 }
2411 return -ENODATA;
2412 }
2413 case BTF_KIND_ENUM:
2414 case BTF_KIND_ENUM64:
2415 err = btf_dump_get_enum_value(d, t, data, id, &value);
2416 if (err)
2417 return err;
2418 if (value == 0)
2419 return -ENODATA;
2420 return 0;
2421 default:
2422 return 0;
2423 }
2424 }
2425
2426 /* 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)2427 static int btf_dump_dump_type_data(struct btf_dump *d,
2428 const char *fname,
2429 const struct btf_type *t,
2430 __u32 id,
2431 const void *data,
2432 __u8 bits_offset,
2433 __u8 bit_sz)
2434 {
2435 int size, err = 0;
2436
2437 size = btf_dump_type_data_check_overflow(d, t, id, data, bits_offset, bit_sz);
2438 if (size < 0)
2439 return size;
2440 err = btf_dump_type_data_check_zero(d, t, id, data, bits_offset, bit_sz);
2441 if (err) {
2442 /* zeroed data is expected and not an error, so simply skip
2443 * dumping such data. Record other errors however.
2444 */
2445 if (err == -ENODATA)
2446 return size;
2447 return err;
2448 }
2449 btf_dump_data_pfx(d);
2450
2451 if (!d->typed_dump->skip_names) {
2452 if (fname && strlen(fname) > 0)
2453 btf_dump_printf(d, ".%s = ", fname);
2454 btf_dump_emit_type_cast(d, id, true);
2455 }
2456
2457 t = skip_mods_and_typedefs(d->btf, id, NULL);
2458
2459 switch (btf_kind(t)) {
2460 case BTF_KIND_UNKN:
2461 case BTF_KIND_FWD:
2462 case BTF_KIND_FUNC:
2463 case BTF_KIND_FUNC_PROTO:
2464 case BTF_KIND_DECL_TAG:
2465 err = btf_dump_unsupported_data(d, t, id);
2466 break;
2467 case BTF_KIND_INT:
2468 if (bit_sz)
2469 err = btf_dump_bitfield_data(d, t, data, bits_offset, bit_sz);
2470 else
2471 err = btf_dump_int_data(d, t, id, data, bits_offset);
2472 break;
2473 case BTF_KIND_FLOAT:
2474 err = btf_dump_float_data(d, t, id, data);
2475 break;
2476 case BTF_KIND_PTR:
2477 err = btf_dump_ptr_data(d, t, id, data);
2478 break;
2479 case BTF_KIND_ARRAY:
2480 err = btf_dump_array_data(d, t, id, data);
2481 break;
2482 case BTF_KIND_STRUCT:
2483 case BTF_KIND_UNION:
2484 err = btf_dump_struct_data(d, t, id, data);
2485 break;
2486 case BTF_KIND_ENUM:
2487 case BTF_KIND_ENUM64:
2488 /* handle bitfield and int enum values */
2489 if (bit_sz) {
2490 __u64 print_num;
2491 __s64 enum_val;
2492
2493 err = btf_dump_get_bitfield_value(d, t, data, bits_offset, bit_sz,
2494 &print_num);
2495 if (err)
2496 break;
2497 enum_val = (__s64)print_num;
2498 err = btf_dump_enum_data(d, t, id, &enum_val);
2499 } else
2500 err = btf_dump_enum_data(d, t, id, data);
2501 break;
2502 case BTF_KIND_VAR:
2503 err = btf_dump_var_data(d, t, id, data);
2504 break;
2505 case BTF_KIND_DATASEC:
2506 err = btf_dump_datasec_data(d, t, id, data);
2507 break;
2508 default:
2509 pr_warn("unexpected kind [%u] for id [%u]\n",
2510 BTF_INFO_KIND(t->info), id);
2511 return -EINVAL;
2512 }
2513 if (err < 0)
2514 return err;
2515 return size;
2516 }
2517
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)2518 int btf_dump__dump_type_data(struct btf_dump *d, __u32 id,
2519 const void *data, size_t data_sz,
2520 const struct btf_dump_type_data_opts *opts)
2521 {
2522 struct btf_dump_data typed_dump = {};
2523 const struct btf_type *t;
2524 int ret;
2525
2526 if (!OPTS_VALID(opts, btf_dump_type_data_opts))
2527 return libbpf_err(-EINVAL);
2528
2529 t = btf__type_by_id(d->btf, id);
2530 if (!t)
2531 return libbpf_err(-ENOENT);
2532
2533 d->typed_dump = &typed_dump;
2534 d->typed_dump->data_end = data + data_sz;
2535 d->typed_dump->indent_lvl = OPTS_GET(opts, indent_level, 0);
2536
2537 /* default indent string is a tab */
2538 if (!OPTS_GET(opts, indent_str, NULL))
2539 d->typed_dump->indent_str[0] = '\t';
2540 else
2541 libbpf_strlcpy(d->typed_dump->indent_str, opts->indent_str,
2542 sizeof(d->typed_dump->indent_str));
2543
2544 d->typed_dump->compact = OPTS_GET(opts, compact, false);
2545 d->typed_dump->skip_names = OPTS_GET(opts, skip_names, false);
2546 d->typed_dump->emit_zeroes = OPTS_GET(opts, emit_zeroes, false);
2547
2548 ret = btf_dump_dump_type_data(d, NULL, t, id, data, 0, 0);
2549
2550 d->typed_dump = NULL;
2551
2552 return libbpf_err(ret);
2553 }
2554