1 /* Most of this is Copyright Dima Kogan <dima@secretsauce.net>
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>.
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.
12 #include <elfutils/libdwfl.h>
19 #include "prototype.h"
24 #include "lens_enum.h"
32 //#define DUMP_PROTOTYPES
35 #define complain( die, format, ... ) \
36 fprintf(stderr, "%s() die '%s' @ 0x%lx: " format "\n", \
37 __func__, dwarf_diename(die), dwarf_dieoffset(die), \
40 #define complain( die, format, ... )
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;
49 static bool get_type(struct arg_type_info** info, Dwarf_Die* type_die);
53 static bool _dump_dwarf_tree(Dwarf_Die* die, int indent)
56 fprintf(stderr, "%*sprocessing unit: 0x%02x/'%s'\n", indent*4, "",
57 dwarf_tag(die), dwarf_diename(die));
60 if (dwarf_child(die, &child) == 0) {
61 if (!_dump_dwarf_tree(&child, indent+1))
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
74 static bool dump_dwarf_tree(Dwarf_Die* die)
76 return _dump_dwarf_tree( die, 0 );
80 #ifdef DUMP_PROTOTYPES
81 static bool _dump_ltrace_tree(const struct arg_type_info* info, int indent)
84 fprintf(stderr, "%*s%p ...\n", indent*4, "", (void*)info);
89 fprintf(stderr, "%*s%p NULL\n", indent*4, "", (void*)info);
95 fprintf(stderr, "%*s%p void\n", indent*4, "", (void*)info);
107 fprintf(stderr, "%*s%p base\n", indent*4, "", (void*)info);
111 fprintf(stderr, "%*s%p array. elements not printed\n", indent*4, "",
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 );
121 fprintf(stderr, "%*s%p struct...\n", indent*4, "", (void*)info);
124 struct arg_type_info *info;
126 }* elements = (struct struct_field*)info->u.entries.data;
128 for(i=0; i<info->u.entries.size; i++)
129 _dump_ltrace_tree( elements[i].info, indent+1 );
133 fprintf(stderr, "%*s%p unknown type\n", indent*4, "", (void*)info);
140 static bool dump_ltrace_tree(const struct arg_type_info* info)
142 return _dump_ltrace_tree( info, 0 );
148 static uint64_t attr_numeric(Dwarf_Die *die, uint32_t name)
150 Dwarf_Attribute attr;
153 if (dwarf_attr(die, name, &attr) == NULL)
156 form = dwarf_whatform(&attr);
161 if (dwarf_formaddr(&attr, &addr) == 0)
170 case DW_FORM_udata: {
172 if (dwarf_formudata(&attr, &value) == 0)
177 case DW_FORM_flag_present: {
179 if (dwarf_formflag(&attr, &value) == 0)
184 complain(die, "DW_AT_<0x%x>=0x%x", name, form);
191 static enum arg_type get_base_type(Dwarf_Die* die)
193 int encoding = attr_numeric(die, DW_AT_encoding);
195 if (encoding == DW_ATE_void )
198 if (encoding == DW_ATE_signed_char || encoding == DW_ATE_unsigned_char )
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)) {
210 return is_signed ? ARGTYPE_SHORT : ARGTYPE_USHORT;
213 return is_signed ? ARGTYPE_INT : ARGTYPE_UINT;
216 return is_signed ? ARGTYPE_LONG : ARGTYPE_ULONG;
224 if (encoding == DW_ATE_float) {
225 switch (attr_numeric(die, DW_AT_byte_size)) {
227 return ARGTYPE_FLOAT;
230 return ARGTYPE_DOUBLE;
233 // things like long doubles. ltrace has no support yet, so I just
239 // Unknown encoding. I just say void
243 static bool get_type_die(Dwarf_Die* type_die, Dwarf_Die* die)
245 Dwarf_Attribute attr;
247 dwarf_attr(die, DW_AT_type, &attr) != NULL &&
248 dwarf_formref_die(&attr, type_die) != NULL;
251 static size_t dwarf_die_hash(const void* x)
253 return *(const Dwarf_Off*)x;
255 static int dwarf_die_eq(const void* a, const void* b)
257 return *(const Dwarf_Off*)a == *(const Dwarf_Off*)b;
260 static bool get_enum(struct arg_type_info* enum_info, Dwarf_Die* parent)
262 enum_info->type = ARGTYPE_INT;
264 struct enum_lens *lens = calloc(1, sizeof(struct enum_lens));
266 complain(parent, "alloc error");
269 lens_init_enum(lens);
270 enum_info->lens = &lens->super;
273 if (dwarf_child(parent, &die) != 0) {
274 // empty enum. we're done
279 complain(&die, "enum element: 0x%02x/'%s'", dwarf_tag(&die),
280 dwarf_diename(&die));
282 if (dwarf_tag(&die) != DW_TAG_enumerator) {
283 complain(&die, "Enums can have ONLY DW_TAG_enumerator elements");
287 if (!dwarf_hasattr(&die, DW_AT_const_value)) {
288 complain(&die, "Enums MUST have DW_AT_const_value values");
292 const char* key = dwarf_diename(&die);
294 complain(&die, "Enums must have a DW_AT_name key");
297 const char* dupkey = strdup(key);
298 if (dupkey == NULL) {
299 complain(&die, "Couldn't duplicate enum key");
303 struct value* value = calloc( 1, sizeof(struct value));
305 complain(&die, "Couldn't alloc enum value");
309 value_init_detached(value, NULL, type_get_simple( ARGTYPE_INT ), 0);
310 value_set_word(value, attr_numeric(&die, DW_AT_const_value));
312 if (lens_enum_add( lens, dupkey, 0, value, 0 )) {
313 complain(&die, "Couldn't add enum element");
317 int res = dwarf_siblingof(&die, &die);
318 if (res == 0) continue; /* sibling exists */
319 if (res < 0) return false; /* error */
320 break; /* no sibling exists */
326 static bool get_array(struct arg_type_info* array_info, Dwarf_Die* parent)
329 if (!get_type_die( &type_die, parent )) {
330 complain( parent, "Array has unknown type" );
334 struct arg_type_info* info;
335 if (!get_type( &info, &type_die )) {
336 complain( parent, "Couldn't figure out array's type" );
341 if (dwarf_child(parent, &subrange) != 0) {
343 "Array must have a DW_TAG_subrange_type child, but has none");
347 Dwarf_Die next_subrange;
348 if (dwarf_siblingof(&subrange, &next_subrange) <= 0) {
350 "Array must have exactly one DW_TAG_subrange_type child");
354 if (dwarf_hasattr(&subrange, DW_AT_lower_bound)) {
355 if (attr_numeric(&subrange, DW_AT_lower_bound) != 0) {
357 "Array subrange has a nonzero lower bound. Don't know what to do");
363 if (!dwarf_hasattr(&subrange, DW_AT_upper_bound)) {
364 // no upper bound is defined. This is probably a variable-width array,
365 // and I don't know how long it is. Let's say 0 to be safe
369 N = attr_numeric(&subrange, DW_AT_upper_bound)+1;
371 // I'm not checking the subrange type. It should be some sort of integer,
372 // and I don't know what it would mean for it to be something else
374 struct value* value = calloc( 1, sizeof(struct value));
376 complain(&subrange, "Couldn't alloc length value");
379 value_init_detached(value, NULL, type_get_simple( ARGTYPE_INT ), 0);
380 value_set_word(value, N );
382 struct expr_node* length = calloc( 1, sizeof(struct expr_node));
383 if (length == NULL) {
384 complain(&subrange, "Couldn't alloc length expr");
387 expr_init_const(length, value);
389 type_init_array(array_info, info, 0, length, 0 );
394 static bool get_structure(struct arg_type_info* struct_info, Dwarf_Die* parent)
396 type_init_struct(struct_info);
399 if (dwarf_child(parent, &die) != 0) {
400 // no elements; we're done
405 complain(&die, "member: 0x%02x", dwarf_tag(&die));
407 if (dwarf_tag(&die) != DW_TAG_member) {
408 complain(&die, "Structure can have ONLY DW_TAG_member");
413 if (!get_type_die( &type_die, &die )) {
414 complain( &die, "Couldn't get type of element");
418 struct arg_type_info* member_info = NULL;
419 if (!get_type( &member_info, &type_die )) {
420 complain(&die, "Couldn't parse type from DWARF data");
423 type_struct_add( struct_info, member_info, 0 );
425 int res = dwarf_siblingof(&die, &die);
426 if (res == 0) continue; /* sibling exists */
427 if (res < 0) return false; /* error */
428 break; /* no sibling exists */
434 // Reads the type in the die into the given structure
435 // Returns true on sucess
436 static bool get_type(struct arg_type_info** info, Dwarf_Die* type_die)
438 Dwarf_Off die_offset = dwarf_dieoffset(type_die);
439 struct arg_type_info** found_type = dict_find(&type_hash, &die_offset );
440 if (found_type != NULL) {
442 complain(type_die, "Read pre-computed type: %p", *info);
448 switch (dwarf_tag(type_die)) {
449 case DW_TAG_base_type:
450 *info = type_get_simple( get_base_type( type_die ));
451 complain(type_die, "Storing base type: %p", *info);
452 dict_insert( &type_hash, &die_offset, info );
455 case DW_TAG_subroutine_type:
456 case DW_TAG_inlined_subroutine:
457 // function pointers are stored as void*. If ltrace tries to dereference
458 // these, it'll get a segfault
459 *info = type_get_simple( ARGTYPE_VOID );
460 complain(type_die, "Storing subroutine type: %p", *info);
461 dict_insert( &type_hash, &die_offset, info );
464 case DW_TAG_pointer_type:
466 if (!get_type_die(&next_die, type_die )) {
467 // the pointed-to type isn't defined, so I report a void*
468 *info = type_get_simple( ARGTYPE_VOID );
469 complain(type_die, "Storing void-pointer type: %p", *info);
470 dict_insert( &type_hash, &die_offset, info );
474 *info = calloc( 1, sizeof(struct arg_type_info));
476 complain(type_die, "alloc error");
479 type_init_pointer(*info, NULL, 0);
481 complain(type_die, "Storing pointer type: %p", *info);
482 dict_insert( &type_hash, &die_offset, info );
483 return get_type( &(*info)->u.ptr_info.info, &next_die );
485 case DW_TAG_structure_type:
486 *info = calloc( 1, sizeof(struct arg_type_info));
488 complain(type_die, "alloc error");
492 complain(type_die, "Storing struct type: %p", *info);
493 dict_insert( &type_hash, &die_offset, info );
494 return get_structure( *info, type_die );
497 case DW_TAG_typedef: ;
498 case DW_TAG_const_type: ;
499 case DW_TAG_volatile_type: ;
500 // Various tags are simply pass-through, so I just keep going
502 if (get_type_die(&next_die, type_die )) {
503 complain(type_die, "Storing const/typedef type: %p", *info);
504 res = get_type( info, &next_die );
506 // no type. Use 'void'. Normally I'd think this is bogus, but stdio
507 // typedefs something to void
508 *info = type_get_simple( ARGTYPE_VOID );
509 complain(type_die, "Storing void type: %p", *info);
512 dict_insert( &type_hash, &die_offset, info );
515 case DW_TAG_enumeration_type:
516 // We have an enumeration. This has type "int", but has a particular
517 // lens to handle the enum
518 *info = calloc( 1, sizeof(struct arg_type_info));
520 complain(type_die, "alloc error");
524 complain(type_die, "Storing enum int: %p", *info);
525 dict_insert( &type_hash, &die_offset, info );
526 return get_enum( *info, type_die );
528 case DW_TAG_array_type:
529 *info = calloc( 1, sizeof(struct arg_type_info));
531 complain(type_die, "alloc error");
535 complain(type_die, "Storing array: %p", *info);
536 dict_insert( &type_hash, &die_offset, info );
537 return get_array( *info, type_die );
539 case DW_TAG_union_type:
540 *info = type_get_simple( ARGTYPE_VOID );
541 complain(type_die, "Storing union-as-void type: %p", *info);
545 complain(type_die, "Unknown type tag 0x%x", dwarf_tag(type_die));
552 static bool get_prototype(struct prototype* proto, Dwarf_Die* subroutine)
554 // First, look at the return type. This is stored in a DW_AT_type tag in the
555 // subroutine DIE. If there is no such tag, this function returns void
556 Dwarf_Die return_type_die;
557 if (!get_type_die(&return_type_die, subroutine )) {
558 proto->return_info = type_get_simple( ARGTYPE_VOID );
559 proto->own_return_info = 0;
561 proto->return_info = calloc( 1, sizeof( struct arg_type_info ));
562 if (proto->return_info == NULL) {
563 complain(subroutine, "Couldn't alloc return type");
566 proto->own_return_info = 0;
568 if (!get_type( &proto->return_info, &return_type_die )) {
569 complain(subroutine, "Couldn't get return type");
575 // Now look at the arguments
577 if (dwarf_child(subroutine, &arg_die) != 0) {
578 // no args. We're done
583 if (dwarf_tag(&arg_die) != DW_TAG_formal_parameter )
584 goto next_prototype_argument;
586 complain(&arg_die, "arg: 0x%02x", dwarf_tag(&arg_die));
589 if (!get_type_die(&type_die, &arg_die )) {
590 complain(&arg_die, "Couldn't get the argument type die");
594 struct arg_type_info* arg_type_info = NULL;
595 if (!get_type( &arg_type_info, &type_die )) {
596 complain(&arg_die, "Couldn't parse arg type from DWARF data");
601 param_init_type(¶m, arg_type_info, 0);
602 if (prototype_push_param(proto, ¶m) <0) {
603 complain(&arg_die, "couldn't add argument to the prototype");
607 #ifdef DUMP_PROTOTYPES
608 fprintf(stderr, "Adding argument:\n");
609 dump_ltrace_tree(arg_type_info);
612 next_prototype_argument: ;
613 int res = dwarf_siblingof(&arg_die, &arg_die);
614 if (res == 0) continue; /* sibling exists */
615 if (res < 0) return false; /* error */
616 break; /* no sibling exists */
622 static bool process_die_compileunit(struct protolib* plib, struct library* lib,
626 if (dwarf_child(parent, &die) != 0) {
627 // no child nodes, so nothing to do
632 if (dwarf_tag(&die) == DW_TAG_subprogram) {
633 const char* function_name = dwarf_diename(&die);
635 complain(&die, "subroutine_type: 0x%02x; function '%s'",
636 dwarf_tag(&die), function_name);
638 struct prototype* proto =
639 protolib_lookup_prototype(plib, function_name, true );
642 complain(&die, "Prototype already exists. Skipping");
646 if (!filter_matches_symbol(options.plt_filter, function_name, lib) &&
647 !filter_matches_symbol(options.static_filter, function_name, lib) &&
648 !filter_matches_symbol(options.export_filter, function_name, lib)) {
649 complain(&die, "Prototype not requested by any filter");
653 proto = malloc(sizeof(struct prototype));
655 complain(&die, "couldn't alloc prototype");
658 prototype_init( proto );
660 if (!get_prototype(proto, &die )) {
661 complain(&die, "couldn't get prototype");
665 protolib_add_prototype(plib, function_name, 0, proto);
669 int res = dwarf_siblingof(&die, &die);
670 if (res == 0) continue; /* sibling exists */
671 if (res < 0) return false; /* error */
672 break; /* no sibling exists */
678 static bool import(struct protolib* plib, struct library* lib, Dwfl* dwfl)
680 dict_init(&type_hash, sizeof(Dwarf_Off), sizeof(struct arg_type_info*),
681 dwarf_die_hash, dwarf_die_eq, NULL );
684 Dwarf_Die* die = NULL;
685 while ((die = dwfl_nextcu(dwfl, die, &bias)) != NULL) {
686 if (dwarf_tag(die) == DW_TAG_compile_unit) {
687 if (!process_die_compileunit(plib, lib, die)) {
688 complain(die, "Error reading compile unit");
693 complain(die, "DW_TAG_compile_unit expected");
699 dict_destroy( &type_hash, NULL, NULL, NULL );
703 bool import_DWARF_prototypes(struct protolib* plib, struct library* lib,
707 plib = protolib_cache_default(&g_protocache, lib->soname, 0);
709 fprintf(stderr, "Error loading protolib %s: %s.\n",
710 lib->soname, strerror(errno));
714 return import(plib, lib, dwfl);
718 - I handle static functions now. Should I? Those do not have DW_AT_external==1
720 - should process existing prototypes to make sure they match
722 - what do function pointers look like? I'm doing void*
726 - all my *allocs leak
728 - protolib_lookup_prototype should look for imports?