449a750052504913a1306f793871ae19e6b1858a
[platform/upstream/ltrace.git] / dwarf_prototypes.c
1 /* Most of this is Copyright Dima Kogan <dima@secretsauce.net>
2  *
3  * Pieces of this were taken from dwarf_prototypes.c in the dwarves project.
4  * Those are Copyright (C) 2008 Arnaldo Carvalho de Melo <acme@redhat.com>.
5  *
6  * This program is free software; you can redistribute it and/or modify it under
7  * the terms of version 2 of the GNU General Public License as published by the
8  * Free Software Foundation.
9  *
10  */
11 #include <stdio.h>
12 #include <elfutils/libdwfl.h>
13 #include <dwarf.h>
14 #include <stdlib.h>
15 #include <errno.h>
16 #include <string.h>
17
18 #include "config.h"
19 #include "prototype.h"
20 #include "type.h"
21 #include "param.h"
22 #include "dict.h"
23 #include "lens.h"
24 #include "lens_enum.h"
25 #include "value.h"
26 #include "expr.h"
27 #include "library.h"
28 #include "options.h"
29 #include "filter.h"
30
31
32 //#define DUMP_PROTOTYPES
33
34 #if 1
35 #define complain( die, format, ... )                                                    \
36         fprintf(stderr, "%s() die '%s' @ 0x%lx: " format "\n",          \
37                         __func__, dwarf_diename(die), dwarf_dieoffset(die),     \
38                         ##__VA_ARGS__ )
39 #else
40 #define complain( die, format, ... )
41 #endif
42
43 // A map from DIE addresses (Dwarf_Off) to type structures (struct
44 // arg_type_info*). This is created and filled in at the start of each import,
45 // and deleted when the import is complete
46 static struct dict type_hash;
47
48
49 static bool get_type(struct arg_type_info** info, Dwarf_Die* type_die);
50
51
52 #if 0
53 static bool _dump_dwarf_tree(Dwarf_Die* die, int indent)
54 {
55     while (1) {
56         fprintf(stderr, "%*sprocessing unit: 0x%02x/'%s'\n", indent*4, "",
57                dwarf_tag(die), dwarf_diename(die));
58
59         Dwarf_Die child;
60         if (dwarf_child(die, &child) == 0) {
61                         if (!_dump_dwarf_tree(&child, indent+1))
62                                 return false;
63         }
64
65         int res = dwarf_siblingof(die, die);
66         if (res == 0 ) continue;     // sibling exists
67         if (res < 0 )  return false; // error
68         break;                       // no sibling exists
69     }
70
71     return true;
72 }
73
74 static bool dump_dwarf_tree(Dwarf_Die* die)
75 {
76     return _dump_dwarf_tree( die, 0 );
77 }
78 #endif
79
80 #ifdef DUMP_PROTOTYPES
81 static bool _dump_ltrace_tree(const struct arg_type_info* info, int indent)
82 {
83         if (indent > 7) {
84                 fprintf(stderr, "%*s%p ...\n", indent*4, "", (void*)info);
85                 return true;
86         }
87
88         if (info == NULL) {
89                 fprintf(stderr, "%*s%p NULL\n", indent*4, "", (void*)info);
90                 return true;
91         }
92
93         switch (info->type) {
94         case ARGTYPE_VOID:
95                 fprintf(stderr, "%*s%p void\n", indent*4, "", (void*)info);
96                 break;
97
98         case ARGTYPE_INT:
99         case ARGTYPE_UINT:
100         case ARGTYPE_LONG:
101         case ARGTYPE_ULONG:
102         case ARGTYPE_CHAR:
103         case ARGTYPE_SHORT:
104         case ARGTYPE_USHORT:
105         case ARGTYPE_FLOAT:
106         case ARGTYPE_DOUBLE:
107                 fprintf(stderr, "%*s%p base\n", indent*4, "", (void*)info);
108                 break;
109
110         case ARGTYPE_ARRAY:
111                 fprintf(stderr, "%*s%p array. elements not printed\n", indent*4, "",
112                                 (void*)info);
113                 break;
114
115         case ARGTYPE_POINTER:
116                 fprintf(stderr, "%*s%p pointer to...\n", indent*4, "", (void*)info);
117                 _dump_ltrace_tree( info->u.ptr_info.info, indent+1 );
118                 break;
119
120         case ARGTYPE_STRUCT:
121                 fprintf(stderr, "%*s%p struct...\n", indent*4, "", (void*)info);
122                 struct struct_field
123                 {
124                         struct arg_type_info *info;
125                         int own_info;
126                 }* elements = (struct struct_field*)info->u.entries.data;
127                 unsigned int i;
128                 for(i=0; i<info->u.entries.size; i++)
129                         _dump_ltrace_tree( elements[i].info, indent+1 );
130                 break;
131
132         default:
133                 fprintf(stderr, "%*s%p unknown type\n", indent*4, "", (void*)info);
134                 return false;;
135         }
136
137         return true;
138 }
139
140 static bool dump_ltrace_tree(const struct arg_type_info* info)
141 {
142         return _dump_ltrace_tree( info, 0 );
143 }
144 #endif
145
146
147
148 static uint64_t attr_numeric(Dwarf_Die *die, uint32_t name)
149 {
150         Dwarf_Attribute attr;
151         uint32_t form;
152
153         if (dwarf_attr(die, name, &attr) == NULL)
154                 return 0;
155
156         form = dwarf_whatform(&attr);
157
158         switch (form) {
159         case DW_FORM_addr: {
160                 Dwarf_Addr addr;
161                 if (dwarf_formaddr(&attr, &addr) == 0)
162                         return addr;
163         }
164                 break;
165         case DW_FORM_data1:
166         case DW_FORM_data2:
167         case DW_FORM_data4:
168         case DW_FORM_data8:
169         case DW_FORM_sdata:
170         case DW_FORM_udata: {
171                 Dwarf_Word value;
172                 if (dwarf_formudata(&attr, &value) == 0)
173                         return value;
174         }
175                 break;
176         case DW_FORM_flag:
177         case DW_FORM_flag_present: {
178                 bool value;
179                 if (dwarf_formflag(&attr, &value) == 0)
180                         return value;
181         }
182                 break;
183         default:
184                 complain(die, "DW_AT_<0x%x>=0x%x", name, form);
185                 break;
186         }
187
188         return 0;
189 }
190
191 static enum arg_type get_base_type(Dwarf_Die* die)
192 {
193         int encoding = attr_numeric(die, DW_AT_encoding);
194
195         if (encoding == DW_ATE_void )
196                 return ARGTYPE_VOID;
197
198         if (encoding == DW_ATE_signed_char || encoding == DW_ATE_unsigned_char )
199                 return ARGTYPE_CHAR;
200
201         if (encoding == DW_ATE_signed   ||
202                 encoding == DW_ATE_unsigned ||
203                 encoding == DW_ATE_boolean) {
204                 bool is_signed = (encoding == DW_ATE_signed);
205                 switch (attr_numeric(die, DW_AT_byte_size)) {
206                 case sizeof(char):
207                         return ARGTYPE_CHAR;
208
209                 case sizeof(short):
210                         return is_signed ? ARGTYPE_SHORT : ARGTYPE_USHORT;
211
212                 case sizeof(int):
213                         return is_signed ? ARGTYPE_INT : ARGTYPE_UINT;
214
215                 case sizeof(long):
216                         return is_signed ? ARGTYPE_LONG : ARGTYPE_ULONG;
217
218                 default:
219                         complain(die, "");
220                         exit(1);
221                 }
222         }
223
224         if (encoding == DW_ATE_float) {
225                 switch (attr_numeric(die, DW_AT_byte_size)) {
226                 case sizeof(float):
227                         return ARGTYPE_FLOAT;
228
229                 case sizeof(double):
230                         return ARGTYPE_DOUBLE;
231
232                 default:
233                         // things like long doubles. ltrace has no support yet, so I just
234                         // say "void"
235                         return ARGTYPE_VOID;
236                 }
237         }
238
239         // Unknown encoding. I just say void
240         complain(die, "Unknown base type. Returning 'void'");
241         return ARGTYPE_VOID;
242 }
243
244 static bool get_type_die(Dwarf_Die* type_die, Dwarf_Die* die)
245 {
246         Dwarf_Attribute attr;
247         return
248                 dwarf_attr(die, DW_AT_type, &attr) != NULL &&
249                 dwarf_formref_die(&attr, type_die) != NULL;
250 }
251
252 static size_t dwarf_die_hash(const void* x)
253 {
254         return *(const Dwarf_Off*)x;
255 }
256 static int dwarf_die_eq(const void* a, const void* b)
257 {
258         return *(const Dwarf_Off*)a == *(const Dwarf_Off*)b;
259 }
260
261 static bool get_enum(struct arg_type_info* enum_info, Dwarf_Die* parent)
262 {
263         enum_info->type = ARGTYPE_INT;
264
265         struct enum_lens *lens = calloc(1, sizeof(struct enum_lens));
266         if (lens == NULL) {
267                 complain(parent, "alloc error");
268                 return false;
269         }
270         lens_init_enum(lens);
271         enum_info->lens = &lens->super;
272
273         Dwarf_Die die;
274         if (dwarf_child(parent, &die) != 0) {
275                 // empty enum. we're done
276                 return true;
277         }
278
279         while(1) {
280                 complain(&die, "enum element: 0x%02x/'%s'", dwarf_tag(&die),
281                                  dwarf_diename(&die));
282
283                 if (dwarf_tag(&die) != DW_TAG_enumerator) {
284                         complain(&die, "Enums can have ONLY DW_TAG_enumerator elements");
285                         return false;
286                 }
287
288                 if (!dwarf_hasattr(&die, DW_AT_const_value)) {
289                         complain(&die, "Enums MUST have DW_AT_const_value values");
290                         return false;
291                 }
292
293                 const char* key = dwarf_diename(&die);
294                 if (key == NULL) {
295                         complain(&die, "Enums must have a DW_AT_name key");
296                         return false;
297                 }
298                 const char* dupkey = strdup(key);
299                 if (dupkey == NULL) {
300                         complain(&die, "Couldn't duplicate enum key");
301                         return false;
302                 }
303
304                 struct value* value = calloc( 1, sizeof(struct value));
305                 if (value == NULL) {
306                         complain(&die, "Couldn't alloc enum value");
307                         return false;
308                 }
309
310                 value_init_detached(value, NULL, type_get_simple( ARGTYPE_INT ), 0);
311                 value_set_word(value, attr_numeric(&die, DW_AT_const_value));
312
313                 if (lens_enum_add( lens, dupkey, 0, value, 0 )) {
314                         complain(&die, "Couldn't add enum element");
315                         return false;
316                 }
317
318                 int res = dwarf_siblingof(&die, &die);
319                 if (res == 0) continue;     /* sibling exists    */
320                 if (res < 0)  return false; /* error             */
321                 break;                      /* no sibling exists */
322         }
323
324         return true;
325 }
326
327 static bool get_array(struct arg_type_info* array_info, Dwarf_Die* parent)
328 {
329         Dwarf_Die type_die;
330         if (!get_type_die( &type_die, parent )) {
331                 complain( parent, "Array has unknown type" );
332                 return false;
333         }
334
335         struct arg_type_info* info;
336         if (!get_type( &info, &type_die )) {
337                 complain( parent, "Couldn't figure out array's type" );
338                 return false;
339         }
340
341         Dwarf_Die subrange;
342         if (dwarf_child(parent, &subrange) != 0) {
343                 complain(parent,
344                                  "Array must have a DW_TAG_subrange_type child, but has none");
345                 return false;
346         }
347
348         Dwarf_Die next_subrange;
349         if (dwarf_siblingof(&subrange, &next_subrange) <= 0) {
350                 complain(parent,
351                                  "Array must have exactly one DW_TAG_subrange_type child");
352                 return false;
353         }
354
355         if (dwarf_hasattr(&subrange, DW_AT_lower_bound)) {
356                 if (attr_numeric(&subrange, DW_AT_lower_bound) != 0) {
357                         complain( parent,
358                                           "Array subrange has a nonzero lower bound. Don't know what to do");
359                         return false;
360                 }
361         }
362
363         int N;
364         if (!dwarf_hasattr(&subrange, DW_AT_upper_bound)) {
365                 // no upper bound is defined. This is probably a variable-width array,
366                 // and I don't know how long it is. Let's say 0 to be safe
367                 N = 0;
368         }
369         else
370                 N = attr_numeric(&subrange, DW_AT_upper_bound)+1;
371
372         // I'm not checking the subrange type. It should be some sort of integer,
373         // and I don't know what it would mean for it to be something else
374
375         struct value* value = calloc( 1, sizeof(struct value));
376         if (value == NULL) {
377                 complain(&subrange, "Couldn't alloc length value");
378                 return false;
379         }
380         value_init_detached(value, NULL, type_get_simple( ARGTYPE_INT ), 0);
381         value_set_word(value, N );
382
383         struct expr_node* length = calloc( 1, sizeof(struct expr_node));
384         if (length == NULL) {
385                 complain(&subrange, "Couldn't alloc length expr");
386                 return false;
387         }
388         expr_init_const(length, value);
389
390         type_init_array(array_info, info, 0, length, 0 );
391
392         return true;
393 }
394
395 static bool get_structure(struct arg_type_info* struct_info, Dwarf_Die* parent)
396 {
397         type_init_struct(struct_info);
398
399         Dwarf_Die die;
400         if (dwarf_child(parent, &die) != 0) {
401                 // no elements; we're done
402                 return true;
403         }
404
405         while(1) {
406                 complain(&die, "member: 0x%02x", dwarf_tag(&die));
407
408                 if (dwarf_tag(&die) != DW_TAG_member) {
409                         complain(&die, "Structure can have ONLY DW_TAG_member");
410                         return false;
411                 }
412
413                 Dwarf_Die type_die;
414                 if (!get_type_die( &type_die, &die )) {
415                         complain( &die, "Couldn't get type of element");
416                         return false;
417                 }
418
419                 struct arg_type_info* member_info = NULL;
420                 if (!get_type( &member_info, &type_die )) {
421                         complain(&die, "Couldn't parse type from DWARF data");
422                         return false;
423                 }
424                 type_struct_add( struct_info, member_info, 0 );
425
426                 int res = dwarf_siblingof(&die, &die);
427                 if (res == 0) continue;     /* sibling exists    */
428                 if (res < 0)  return false; /* error             */
429                 break;                      /* no sibling exists */
430         }
431
432         return true;
433 }
434
435 // Reads the type in the die into the given structure
436 // Returns true on sucess
437 static bool get_type(struct arg_type_info** info, Dwarf_Die* type_die)
438 {
439         Dwarf_Off die_offset = dwarf_dieoffset(type_die);
440         struct arg_type_info** found_type = dict_find(&type_hash, &die_offset );
441         if (found_type != NULL) {
442                 *info = *found_type;
443                 complain(type_die, "Read pre-computed type: %p", *info);
444                 return true;
445         }
446
447         Dwarf_Die next_die;
448
449         switch (dwarf_tag(type_die)) {
450         case DW_TAG_base_type:
451                 *info = type_get_simple( get_base_type( type_die ));
452                 complain(type_die, "Storing base type: %p", *info);
453                 dict_insert( &type_hash, &die_offset, info );
454                 return true;
455
456         case DW_TAG_subroutine_type:
457         case DW_TAG_inlined_subroutine:
458                 // function pointers are stored as void*. If ltrace tries to dereference
459                 // these, it'll get a segfault
460                 *info = type_get_simple( ARGTYPE_VOID );
461                 complain(type_die, "Storing subroutine type: %p", *info);
462                 dict_insert( &type_hash, &die_offset, info );
463                 return true;
464
465         case DW_TAG_pointer_type:
466
467                 if (!get_type_die(&next_die, type_die )) {
468                         // the pointed-to type isn't defined, so I report a void*
469                         *info = type_get_simple( ARGTYPE_VOID );
470                         complain(type_die, "Storing void-pointer type: %p", *info);
471                         dict_insert( &type_hash, &die_offset, info );
472                         return true;
473                 }
474
475                 *info = calloc( 1, sizeof(struct arg_type_info));
476                 if (*info == NULL) {
477                         complain(type_die, "alloc error");
478                         return false;
479                 }
480                 type_init_pointer(*info, NULL, 0);
481
482                 complain(type_die, "Storing pointer type: %p", *info);
483                 dict_insert( &type_hash, &die_offset, info );
484                 return get_type( &(*info)->u.ptr_info.info, &next_die );
485
486         case DW_TAG_structure_type:
487                 *info = calloc( 1, sizeof(struct arg_type_info));
488                 if (*info == NULL) {
489                         complain(type_die, "alloc error");
490                         return false;
491                 }
492
493                 complain(type_die, "Storing struct type: %p", *info);
494                 dict_insert( &type_hash, &die_offset, info );
495                 return get_structure( *info, type_die );
496
497
498         case DW_TAG_typedef: ;
499         case DW_TAG_const_type: ;
500         case DW_TAG_volatile_type: ;
501                 // Various tags are simply pass-through, so I just keep going
502                 bool res = true;
503                 if (get_type_die(&next_die, type_die )) {
504                         complain(type_die, "Storing const/typedef type: %p", *info);
505                         res = get_type( info, &next_die );
506                 } else {
507                         // no type. Use 'void'. Normally I'd think this is bogus, but stdio
508                         // typedefs something to void
509                         *info = type_get_simple( ARGTYPE_VOID );
510                         complain(type_die, "Storing void type: %p", *info);
511                 }
512                 if (res )
513                         dict_insert( &type_hash, &die_offset, info );
514                 return res;
515
516         case DW_TAG_enumeration_type:
517                 // We have an enumeration. This has type "int", but has a particular
518                 // lens to handle the enum
519                 *info = calloc( 1, sizeof(struct arg_type_info));
520                 if (*info == NULL) {
521                         complain(type_die, "alloc error");
522                         return false;
523                 }
524
525                 complain(type_die, "Storing enum int: %p", *info);
526                 dict_insert( &type_hash, &die_offset, info );
527                 return get_enum( *info, type_die );
528
529         case DW_TAG_array_type:
530                 *info = calloc( 1, sizeof(struct arg_type_info));
531                 if (*info == NULL) {
532                         complain(type_die, "alloc error");
533                         return false;
534                 }
535
536                 complain(type_die, "Storing array: %p", *info);
537                 dict_insert( &type_hash, &die_offset, info );
538                 return get_array( *info, type_die );
539
540         case DW_TAG_union_type:
541                 *info = type_get_simple( ARGTYPE_VOID );
542                 complain(type_die, "Storing union-as-void type: %p", *info);
543                 return true;
544
545         default:
546                 complain(type_die, "Unknown type tag 0x%x", dwarf_tag(type_die));
547                 break;
548         }
549
550         return false;
551 }
552
553 static bool get_prototype(struct prototype* proto, Dwarf_Die* subroutine)
554 {
555         // First, look at the return type. This is stored in a DW_AT_type tag in the
556         // subroutine DIE. If there is no such tag, this function returns void
557         Dwarf_Die return_type_die;
558         if (!get_type_die(&return_type_die, subroutine )) {
559                 proto->return_info = type_get_simple( ARGTYPE_VOID );
560                 proto->own_return_info = 0;
561         } else {
562                 proto->return_info = calloc( 1, sizeof( struct arg_type_info ));
563                 if (proto->return_info == NULL) {
564                         complain(subroutine, "Couldn't alloc return type");
565                         return false;
566                 }
567                 proto->own_return_info = 0;
568
569                 if (!get_type( &proto->return_info, &return_type_die )) {
570                         complain(subroutine, "Couldn't get return type");
571                         return false;
572                 }
573         }
574
575
576         // Now look at the arguments
577         Dwarf_Die arg_die;
578         if (dwarf_child(subroutine, &arg_die) != 0) {
579                 // no args. We're done
580                 return true;
581         }
582
583         while(1) {
584                 if (dwarf_tag(&arg_die) != DW_TAG_formal_parameter )
585                         goto next_prototype_argument;
586
587                 complain(&arg_die, "arg: 0x%02x", dwarf_tag(&arg_die));
588
589                 Dwarf_Die type_die;
590                 if (!get_type_die(&type_die, &arg_die )) {
591                         complain(&arg_die, "Couldn't get the argument type die");
592                         return false;
593                 }
594
595                 struct arg_type_info* arg_type_info = NULL;
596                 if (!get_type( &arg_type_info, &type_die )) {
597                         complain(&arg_die, "Couldn't parse arg type from DWARF data");
598                         return false;
599                 }
600
601                 struct param param;
602                 param_init_type(&param, arg_type_info, 0);
603                 if (prototype_push_param(proto, &param) <0) {
604                         complain(&arg_die, "couldn't add argument to the prototype");
605                         return false;
606                 }
607
608 #ifdef DUMP_PROTOTYPES
609                 fprintf(stderr, "Adding argument:\n");
610                 dump_ltrace_tree(arg_type_info);
611 #endif
612
613         next_prototype_argument: ;
614                 int res = dwarf_siblingof(&arg_die, &arg_die);
615                 if (res == 0) continue;     /* sibling exists    */
616                 if (res < 0)  return false; /* error             */
617                 break;                      /* no sibling exists */
618         }
619
620         return true;
621 }
622
623 static bool process_die_compileunit(struct protolib* plib, struct library* lib,
624                                                                         Dwarf_Die* parent)
625 {
626         Dwarf_Die die;
627         if (dwarf_child(parent, &die) != 0) {
628                 // no child nodes, so nothing to do
629                 return true;
630         }
631
632         while (1) {
633                 if (dwarf_tag(&die) == DW_TAG_subprogram) {
634                         const char* function_name = dwarf_diename(&die);
635
636                         complain(&die, "subroutine_type: 0x%02x; function '%s'",
637                                          dwarf_tag(&die), function_name);
638
639                         struct prototype* proto =
640                                 protolib_lookup_prototype(plib, function_name, true );
641
642                         if (proto != NULL) {
643                                 complain(&die, "Prototype already exists. Skipping");
644                                 goto next_prototype;
645                         }
646
647                         if (!filter_matches_symbol(options.plt_filter,    function_name, lib) &&
648                                 !filter_matches_symbol(options.static_filter, function_name, lib) &&
649                                 !filter_matches_symbol(options.export_filter, function_name, lib)) {
650                                 complain(&die, "Prototype not requested by any filter");
651                                 goto next_prototype;
652                         }
653
654                         proto = malloc(sizeof(struct prototype));
655                         if (proto == NULL) {
656                                 complain(&die, "couldn't alloc prototype");
657                                 return false;
658                         }
659                         prototype_init( proto );
660
661                         if (!get_prototype(proto, &die )) {
662                                 complain(&die, "couldn't get prototype");
663                                 return false;
664                         }
665
666                         protolib_add_prototype(plib, function_name, 0, proto);
667                 }
668
669                 next_prototype:;
670                 int res = dwarf_siblingof(&die, &die);
671                 if (res == 0) continue;     /* sibling exists    */
672                 if (res < 0)  return false; /* error             */
673                 break;                      /* no sibling exists */
674         }
675
676         return true;
677 }
678
679 static bool import(struct protolib* plib, struct library* lib, Dwfl* dwfl)
680 {
681         dict_init(&type_hash, sizeof(Dwarf_Off), sizeof(struct arg_type_info*),
682                           dwarf_die_hash, dwarf_die_eq, NULL );
683
684         Dwarf_Addr bias;
685     Dwarf_Die* die = NULL;
686     while ((die = dwfl_nextcu(dwfl, die, &bias)) != NULL) {
687         if (dwarf_tag(die) == DW_TAG_compile_unit) {
688             if (!process_die_compileunit(plib, lib, die)) {
689                 complain(die, "Error reading compile unit");
690                                 exit(1);
691                                 return false;
692             }
693         } else {
694             complain(die, "DW_TAG_compile_unit expected");
695                         exit(1);
696             return false;
697         }
698     }
699
700         dict_destroy( &type_hash, NULL, NULL, NULL );
701         return true;
702 }
703
704 bool import_DWARF_prototypes(struct protolib* plib, struct library* lib,
705                                                          Dwfl *dwfl)
706 {
707         if (plib == NULL) {
708                 plib = protolib_cache_default(&g_protocache, lib->soname, 0);
709                 if (plib == NULL) {
710                         fprintf(stderr, "Error loading protolib %s: %s.\n",
711                                         lib->soname, strerror(errno));
712                 }
713         }
714
715         return import(plib, lib, dwfl);
716 }
717
718 /*
719 - I handle static functions now. Should I? Those do not have DW_AT_external==1
720
721 - should process existing prototypes to make sure they match
722
723 - what do function pointers look like? I'm doing void*
724
725 - unions
726
727 - all my *allocs leak
728
729 - protolib_lookup_prototype should look for imports?
730
731 */