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