Sweden-Number/debugger/msc.c

2409 lines
61 KiB
C

/*
* File msc.c - read VC++ debug information from COFF and eventually
* from PDB files.
*
* Copyright (C) 1996, Eric Youngdale.
*
* Note - this handles reading debug information for 32 bit applications
* that run under Windows-NT for example. I doubt that this would work well
* for 16 bit applications, but I don't think it really matters since the
* file format is different, and we should never get in here in such cases.
*
* TODO:
* Get 16 bit CV stuff working.
* Add symbol size to internal symbol table.
*/
#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <sys/mman.h>
#include <fcntl.h>
#include <sys/stat.h>
#include <limits.h>
#include <string.h>
#include <unistd.h>
#ifndef PATH_MAX
#define PATH_MAX _MAX_PATH
#endif
#include "win.h"
#include "pe_image.h"
#include "peexe.h"
#include "debugger.h"
#include "peexe.h"
#include "xmalloc.h"
/*
* This is an index we use to keep track of the debug information
* when we have multiple sources. We use the same database to also
* allow us to do an 'info shared' type of deal, and we use the index
* to eliminate duplicates.
*/
static int DEBUG_next_index = 0;
union any_size
{
char * c;
short * s;
int * i;
unsigned int * ui;
};
/*
* This is a convenience structure used to map portions of the
* line number table.
*/
struct startend
{
unsigned int start;
unsigned int end;
};
/*
* This is how we reference the various record types.
*/
union codeview_symbol
{
struct
{
short int len;
short int id;
} generic;
struct
{
short int len;
short int id;
unsigned int offset;
unsigned short seg;
unsigned short symtype;
unsigned char namelen;
unsigned char name[1];
} data;
struct
{
short int len;
short int id;
unsigned int pparent;
unsigned int pend;
unsigned int next;
unsigned int offset;
unsigned short segment;
unsigned short thunk_len;
unsigned char thtype;
unsigned char namelen;
unsigned char name[1];
} thunk;
struct
{
short int len;
short int id;
unsigned int pparent;
unsigned int pend;
unsigned int next;
unsigned int proc_len;
unsigned int debug_start;
unsigned int debug_end;
unsigned int offset;
unsigned short segment;
unsigned short proctype;
unsigned char flags;
unsigned char namelen;
unsigned char name[1];
} proc;
struct
{
short int len; /* Total length of this entry */
short int id; /* Always S_BPREL32 */
unsigned int offset; /* Stack offset relative to BP */
unsigned short symtype;
unsigned char namelen;
unsigned char name[1];
} stack;
};
union codeview_type
{
struct
{
short int len;
short int id;
} generic;
struct
{
short int len;
short int id;
short int attribute;
short int datatype;
unsigned char variant[1];
} pointer;
struct
{
short int len;
short int id;
unsigned char nbits;
unsigned char bitoff;
unsigned short type;
} bitfield;
struct
{
short int len;
short int id;
short int elemtype;
short int idxtype;
unsigned char arrlen;
unsigned char namelen;
unsigned char name[1];
} array;
struct
{
short int len;
short int id;
short int n_element;
short int fieldlist;
short int property;
short int derived;
short int vshape;
unsigned short structlen;
unsigned char namelen;
unsigned char name[1];
} structure;
struct
{
short int len;
short int id;
short int count;
short int field;
short int property;
unsigned short un_len;
unsigned char namelen;
unsigned char name[1];
} t_union;
struct
{
short int len;
short int id;
short int count;
short int type;
short int field;
short int property;
unsigned char namelen;
unsigned char name[1];
} enumeration;
struct
{
short int id;
short int attribute;
unsigned short int value;
unsigned char namelen;
unsigned char name[1];
} enumerate;
struct
{
short int id;
short int type;
short int attribute;
unsigned short int offset;
unsigned char namelen;
unsigned char name[1];
} member;
struct
{
short int len;
short int id;
short int count;
short int type;
short int field;
short int property;
unsigned char namelen;
unsigned char name[1];
} fieldlist;
};
#define S_BPREL32 0x200
#define S_LDATA32 0x201
#define S_GDATA32 0x202
#define S_PUB32 0x203
#define S_LPROC32 0x204
#define S_GPROC32 0x205
#define S_THUNK32 0x206
#define S_BLOCK32 0x207
#define S_WITH32 0x208
#define S_LABEL32 0x209
#define S_PROCREF 0x400
#define S_DATAREF 0x401
#define S_ALIGN 0x402
#define S_UNKNOWN 0x403
/*
* This covers the basic datatypes that VC++ seems to be using these days.
* 32 bit mode only. There are additional numbers for the pointers in 16
* bit mode. There are many other types listed in the documents, but these
* are apparently not used by the compiler, or represent pointer types
* that are not used.
*/
#define T_NOTYPE 0x0000 /* Notype */
#define T_ABS 0x0001 /* Abs */
#define T_VOID 0x0003 /* Void */
#define T_CHAR 0x0010 /* signed char */
#define T_SHORT 0x0011 /* short */
#define T_LONG 0x0012 /* long */
#define T_QUAD 0x0013 /* long long */
#define T_UCHAR 0x0020 /* unsigned char */
#define T_USHORT 0x0021 /* unsigned short */
#define T_ULONG 0x0022 /* unsigned long */
#define T_UQUAD 0x0023 /* unsigned long long */
#define T_REAL32 0x0040 /* float */
#define T_REAL64 0x0041 /* double */
#define T_RCHAR 0x0070 /* real char */
#define T_WCHAR 0x0071 /* wide char */
#define T_INT4 0x0074 /* int */
#define T_UINT4 0x0075 /* unsigned int */
#define T_32PVOID 0x0403 /* 32 bit near pointer to void */
#define T_32PCHAR 0x0410 /* 16:32 near pointer to signed char */
#define T_32PSHORT 0x0411 /* 16:32 near pointer to short */
#define T_32PLONG 0x0412 /* 16:32 near pointer to int */
#define T_32PQUAD 0x0413 /* 16:32 near pointer to long long */
#define T_32PUCHAR 0x0420 /* 16:32 near pointer to unsigned char */
#define T_32PUSHORT 0x0421 /* 16:32 near pointer to unsigned short */
#define T_32PULONG 0x0422 /* 16:32 near pointer to unsigned int */
#define T_32PUQUAD 0x0423 /* 16:32 near pointer to long long */
#define T_32PREAL32 0x0440 /* 16:32 near pointer to float */
#define T_32PREAL64 0x0441 /* 16:32 near pointer to float */
#define T_32PRCHAR 0x0470 /* 16:32 near pointer to real char */
#define T_32PWCHAR 0x0471 /* 16:32 near pointer to real char */
#define T_32PINT4 0x0474 /* 16:32 near pointer to int */
#define T_32PUINT4 0x0475 /* 16:32 near pointer to unsigned int */
#define LF_MODIFIER 0x1
#define LF_POINTER 0x2
#define LF_ARRAY 0x3
#define LF_CLASS 0x4
#define LF_STRUCTURE 0x5
#define LF_UNION 0x6
#define LF_ENUMERATION 0x7
#define LF_PROCEDURE 0x8
#define LF_MFUNCTION 0x9
#define LF_VTSHAPE 0xa
#define LF_BARRAY 0xd
#define LF_DIMARRAY 0x11
#define LF_VFTPATH 0x12
#define LF_SKIP 0x200
#define LF_ARGLIST 0x201
#define LF_FIELDLIST 0x204
#define LF_DERIVED 0x205
#define LF_BITFIELD 0x206
#define LF_BCLASS 0x400
#define LF_VBCLASS 0x401
#define LF_IVBCLASS 0x402
#define LF_ENUMERATE 0x403
#define LF_FRIENDFCN 0x404
#define LF_INDEX 0x405
#define LF_MEMBER 0x406
#define LF_STMEMBER 0x407
#define LF_METHOD 0x408
#define LF_NESTEDTYPE 0x409
#define LF_VFUNCTAB 0x40a
#define LF_FRIENDCLS 0x40b
#define LF_ONEMETHOD 0x40c
#define LF_FUNCOFF 0x40d
#define MAX_BUILTIN_TYPES 0x480
static struct datatype * cv_basic_types[MAX_BUILTIN_TYPES];
static int num_cv_defined_types = 0;
static struct datatype **cv_defined_types = NULL;
/*
* For the type CODEVIEW debug directory entries, the debug directory
* points to a structure like this. The cv_name field is the name
* of an external .PDB file.
*/
struct CodeViewDebug
{
char cv_nbtype[8];
unsigned int cv_timestamp;
char cv_unknown[4];
char cv_name[1];
};
struct MiscDebug {
unsigned int DataType;
unsigned int Length;
char Unicode;
char Reserved[3];
char Data[1];
};
/*
* This is the header that the COFF variety of debug header points to.
*/
struct CoffDebug {
unsigned int N_Sym;
unsigned int SymbolOffset;
unsigned int N_Linenum;
unsigned int LinenumberOffset;
unsigned int Unused[4];
};
struct CoffLinenum {
unsigned int VirtualAddr;
unsigned short int Linenum;
};
struct CoffFiles {
unsigned int startaddr;
unsigned int endaddr;
char * filename;
int linetab_offset;
int linecnt;
struct name_hash **entries;
int neps;
int neps_alloc;
};
struct CoffSymbol {
union {
char ShortName[8];
struct {
unsigned int NotLong;
unsigned int StrTaboff;
} Name;
} N;
unsigned int Value;
short SectionNumber;
short Type;
char StorageClass;
unsigned char NumberOfAuxSymbols;
};
struct CoffAuxSection{
unsigned int Length;
unsigned short NumberOfRelocations;
unsigned short NumberOfLinenumbers;
unsigned int CheckSum;
short Number;
char Selection;
} Section;
/*
* These two structures are used in the directory within a .DBG file
* to locate the individual important bits that we might want to see.
*/
struct CV4_DirHead {
short unsigned int dhsize;
short unsigned int desize;
unsigned int ndir;
unsigned int next_offset;
unsigned int flags;
};
struct CV4_DirEnt {
short unsigned int subsect_number;
short unsigned int module_number;
unsigned int offset;
unsigned int size;
};
/*
* These are the values of interest that the subsect_number field takes.
*/
#define sstAlignSym 0x125
#define sstSrcModule 0x127
struct codeview_linetab_hdr
{
unsigned int nline;
unsigned int segno;
unsigned int start;
unsigned int end;
char * sourcefile;
unsigned short * linetab;
unsigned int * offtab;
};
struct codeview_pdb_hdr
{
char ident[44];
unsigned int blocksize; /* Extent size */
unsigned short loc_freelist; /* freelist. */
unsigned short alloc_filesize; /* # extents allocated. */
unsigned int toc_len;
unsigned int unknown;
unsigned short toc_ext[1]; /* array of extent #'s for toc. */
};
/*
* This is our own structure that we use to keep track of the contents
* of a PDB file.
*/
struct file_list
{
int record_len;
int nextents;
short int * extent_list;
unsigned int linetab_offset;
unsigned int linetab_len;
};
/*
* These are the structures that represent how the file table is set up
* within the PDB file.
*/
struct filetab_hdr
{
unsigned short tab1_file;
unsigned short tab2_file;
unsigned short gsym_file;
unsigned short padding;
unsigned int ftab_len;
unsigned int fofftab_len;
unsigned int hash_len;
unsigned int strtab_len;
};
struct file_ent
{
unsigned int reserved1;
unsigned short datasect_segment;
unsigned short reserved2;
unsigned int datasect_offset;
unsigned int datasect_size;
unsigned int datasect_flags;
unsigned short reserved3;
unsigned short index;
unsigned short num6a;
unsigned short file_number;
unsigned int linetab_offset;
unsigned int linetab_len;
unsigned int num9;
unsigned int num10;
unsigned int num11;
unsigned char filename[1];
};
/*
********************************************************************
*/
struct deferred_debug_info
{
struct deferred_debug_info * next;
char * load_addr;
char * module_name;
char * dbg_info;
int dbg_size;
HMODULE32 module;
LPIMAGE_DEBUG_DIRECTORY dbgdir;
LPIMAGE_SECTION_HEADER sectp;
int nsect;
short int dbg_index;
char loaded;
};
struct deferred_debug_info * dbglist = NULL;
/*
* A simple macro that tells us whether a given COFF symbol is a
* function or not.
*/
#define N_TMASK 0x0030
#define IMAGE_SYM_DTYPE_FUNCTION 2
#define N_BTSHFT 4
#define ISFCN(x) (((x) & N_TMASK) == (IMAGE_SYM_DTYPE_FUNCTION << N_BTSHFT))
/*
* This is what we are looking for in the COFF symbols.
*/
#define IMAGE_SYM_CLASS_EXTERNAL 0x2
#define IMAGE_SYM_CLASS_STATIC 0x3
#define IMAGE_SYM_CLASS_FILE 0x67
static
struct datatype * DEBUG_GetCVType(int typeno)
{
struct datatype * dt = NULL;
/*
* Convert Codeview type numbers into something we can grok internally.
* Numbers < 0x1000 are all fixed builtin types. Numbers from 0x1000 and
* up are all user defined (structs, etc).
*/
if( typeno < 0x1000 )
{
if( typeno < MAX_BUILTIN_TYPES )
{
dt = cv_basic_types[typeno];
}
}
else
{
if( typeno - 0x1000 < num_cv_defined_types )
{
dt = cv_defined_types[typeno - 0x1000];
}
}
return dt;
}
static int
DEBUG_ParseTypeTable(char * table, int len)
{
int arr_max;
int curr_type;
enum debug_type fieldtype;
int elem_size;
union any_size ptr;
union any_size ptr2;
struct datatype * subtype;
char symname[256];
union codeview_type * type;
union codeview_type * type2;
struct datatype * typeptr;
curr_type = 0x1000;
ptr = (union any_size) (table + 16);
while( ptr.c - table < len )
{
type = (union codeview_type *) ptr.c;
if( curr_type - 0x1000 >= num_cv_defined_types )
{
num_cv_defined_types += 0x100;
cv_defined_types = (struct datatype **) realloc(cv_defined_types,
num_cv_defined_types * sizeof(struct datatype *));
memset(cv_defined_types + num_cv_defined_types - 0x100,
0,
0x100 * sizeof(struct datatype *));
if( cv_defined_types == NULL )
{
return FALSE;
}
}
switch(type->generic.id)
{
case LF_POINTER:
cv_defined_types[curr_type - 0x1000] =
DEBUG_FindOrMakePointerType(DEBUG_GetCVType(type->pointer.datatype));
break;
case LF_ARRAY:
if( type->array.arrlen >= 0x8000 )
{
/*
* This is a numeric leaf, I am too lazy to handle this right
* now.
*/
fprintf(stderr, "Ignoring large numberic leaf.\n");
break;
}
if( type->array.namelen != 0 )
{
memset(symname, 0, sizeof(symname));
memcpy(symname, type->array.name, type->array.namelen);
typeptr = DEBUG_NewDataType(ARRAY, symname);
}
else
{
typeptr = DEBUG_NewDataType(ARRAY, NULL);
}
cv_defined_types[curr_type - 0x1000] = typeptr;
subtype = DEBUG_GetCVType(type->array.elemtype);
if( (subtype == NULL)
|| (elem_size = DEBUG_GetObjectSize(subtype)) == 0 )
{
arr_max = 0;
}
else
{
arr_max = type->array.arrlen / DEBUG_GetObjectSize(subtype);
}
DEBUG_SetArrayParams(typeptr, 0, arr_max, subtype);
break;
case LF_FIELDLIST:
/*
* This is where the basic list of fields is defined for
* structures and classes.
*
* First, we need to look ahead and see whether we are building
* a fieldlist for an enum or a struct.
*/
ptr2.i = ptr.i + 1;
type2 = (union codeview_type *) ptr2.c;
if( type2->member.id == LF_MEMBER )
{
typeptr = DEBUG_NewDataType(STRUCT, NULL);
fieldtype = STRUCT;
}
else if( type2->member.id == LF_ENUMERATE )
{
typeptr = DEBUG_NewDataType(ENUM, NULL);
fieldtype = ENUM;
}
else
{
break;
}
cv_defined_types[curr_type - 0x1000] = typeptr;
while( ptr2.c < (ptr.c + ((type->generic.len + 3) & ~3)) )
{
type2 = (union codeview_type *) ptr2.c;
if( type2->member.id == LF_MEMBER && fieldtype == STRUCT )
{
memset(symname, 0, sizeof(symname));
memcpy(symname, type2->member.name, type2->member.namelen);
subtype = DEBUG_GetCVType(type2->member.type);
elem_size = 0;
if( subtype != NULL )
{
elem_size = DEBUG_GetObjectSize(subtype);
}
if( type2->member.offset >= 0x8000 )
{
/*
* This is a numeric leaf, I am too lazy to handle this right
* now.
*/
fprintf(stderr, "Ignoring large numberic leaf.\n");
}
else
{
DEBUG_AddStructElement(typeptr, symname, subtype,
type2->member.offset << 3,
elem_size << 3);
}
}
else if( type2->member.id == LF_ENUMERATE && fieldtype == ENUM )
{
memset(symname, 0, sizeof(symname));
memcpy(symname, type2->enumerate.name, type2->enumerate.namelen);
if( type2->enumerate.value >= 0x8000 )
{
/*
* This is a numeric leaf, I am too lazy to handle this right
* now.
*/
fprintf(stderr, "Ignoring large numberic leaf.\n");
}
else
{
DEBUG_AddStructElement(typeptr, symname, NULL,
type2->enumerate.value, 0);
}
}
else
{
/*
* Something else I have never seen before. Either wrong type of
* object in the fieldlist, or some other problem which I wouldn't
* really know how to handle until it came up.
*/
fprintf(stderr, "Unexpected entry in fieldlist\n");
break;
}
ptr2.c += ((type2->member.namelen + 9 + 3) & ~3);
}
break;
case LF_STRUCTURE:
case LF_CLASS:
if( type->structure.structlen >= 0x8000 )
{
/*
* This is a numeric leaf, I am too lazy to handle this right
* now.
*/
fprintf(stderr, "Ignoring large numberic leaf.\n");
break;
}
memset(symname, 0, sizeof(symname));
memcpy(symname, type->structure.name, type->structure.namelen);
if( strcmp(symname, "__unnamed") == 0 )
{
typeptr = DEBUG_NewDataType(STRUCT, NULL);
}
else
{
typeptr = DEBUG_NewDataType(STRUCT, symname);
}
cv_defined_types[curr_type - 0x1000] = typeptr;
/*
* Now copy the relevant bits from the fieldlist that we specified.
*/
subtype = DEBUG_GetCVType(type->structure.fieldlist);
if( subtype != NULL )
{
DEBUG_SetStructSize(typeptr, type->structure.structlen);
DEBUG_CopyFieldlist(typeptr, subtype);
}
break;
case LF_UNION:
if( type->t_union.un_len >= 0x8000 )
{
/*
* This is a numeric leaf, I am too lazy to handle this right
* now.
*/
fprintf(stderr, "Ignoring large numberic leaf.\n");
break;
}
memset(symname, 0, sizeof(symname));
memcpy(symname, type->t_union.name, type->t_union.namelen);
if( strcmp(symname, "__unnamed") == 0 )
{
typeptr = DEBUG_NewDataType(STRUCT, NULL);
}
else
{
typeptr = DEBUG_NewDataType(STRUCT, symname);
}
cv_defined_types[curr_type - 0x1000] = typeptr;
/*
* Now copy the relevant bits from the fieldlist that we specified.
*/
subtype = DEBUG_GetCVType(type->t_union.field);
if( subtype != NULL )
{
DEBUG_SetStructSize(typeptr, type->t_union.un_len);
DEBUG_CopyFieldlist(typeptr, subtype);
}
break;
case LF_BITFIELD:
typeptr = DEBUG_NewDataType(BITFIELD, NULL);
cv_defined_types[curr_type - 0x1000] = typeptr;
DEBUG_SetBitfieldParams(typeptr, type->bitfield.bitoff,
type->bitfield.nbits,
DEBUG_GetCVType(type->bitfield.type));
break;
case LF_ENUMERATION:
memset(symname, 0, sizeof(symname));
memcpy(symname, type->enumeration.name, type->enumeration.namelen);
typeptr = DEBUG_NewDataType(ENUM, symname);
cv_defined_types[curr_type - 0x1000] = typeptr;
/*
* Now copy the relevant bits from the fieldlist that we specified.
*/
subtype = DEBUG_GetCVType(type->enumeration.field);
if( subtype != NULL )
{
DEBUG_CopyFieldlist(typeptr, subtype);
}
break;
case LF_DIMARRAY:
default:
break;
}
curr_type++;
ptr.c += (type->generic.len + 3) & ~3;
}
return TRUE;
}
void
DEBUG_InitCVDataTypes()
{
/*
* These are the common builtin types that are used by VC++.
*/
cv_basic_types[T_NOTYPE] = NULL;
cv_basic_types[T_ABS] = NULL;
cv_basic_types[T_VOID] = DEBUG_NewDataType(BASIC, "void");
cv_basic_types[T_CHAR] = DEBUG_NewDataType(BASIC, "char");
cv_basic_types[T_SHORT] = DEBUG_NewDataType(BASIC, "short int");
cv_basic_types[T_LONG] = DEBUG_NewDataType(BASIC, "long int");
cv_basic_types[T_QUAD] = DEBUG_NewDataType(BASIC, "long long int");
cv_basic_types[T_UCHAR] = DEBUG_NewDataType(BASIC, "unsigned char");
cv_basic_types[T_USHORT] = DEBUG_NewDataType(BASIC, "short unsigned int");
cv_basic_types[T_ULONG] = DEBUG_NewDataType(BASIC, "long unsigned int");
cv_basic_types[T_UQUAD] = DEBUG_NewDataType(BASIC, "long long unsigned int");
cv_basic_types[T_REAL32] = DEBUG_NewDataType(BASIC, "float");
cv_basic_types[T_REAL64] = DEBUG_NewDataType(BASIC, "double");
cv_basic_types[T_RCHAR] = DEBUG_NewDataType(BASIC, "char");
cv_basic_types[T_WCHAR] = DEBUG_NewDataType(BASIC, "short");
cv_basic_types[T_INT4] = DEBUG_NewDataType(BASIC, "int");
cv_basic_types[T_UINT4] = DEBUG_NewDataType(BASIC, "unsigned int");
cv_basic_types[T_32PVOID] = DEBUG_FindOrMakePointerType(cv_basic_types[T_VOID]);
cv_basic_types[T_32PCHAR] = DEBUG_FindOrMakePointerType(cv_basic_types[T_CHAR]);
cv_basic_types[T_32PSHORT] = DEBUG_FindOrMakePointerType(cv_basic_types[T_SHORT]);
cv_basic_types[T_32PLONG] = DEBUG_FindOrMakePointerType(cv_basic_types[T_LONG]);
cv_basic_types[T_32PQUAD] = DEBUG_FindOrMakePointerType(cv_basic_types[T_QUAD]);
cv_basic_types[T_32PUCHAR] = DEBUG_FindOrMakePointerType(cv_basic_types[T_UCHAR]);
cv_basic_types[T_32PUSHORT] = DEBUG_FindOrMakePointerType(cv_basic_types[T_USHORT]);
cv_basic_types[T_32PULONG] = DEBUG_FindOrMakePointerType(cv_basic_types[T_ULONG]);
cv_basic_types[T_32PUQUAD] = DEBUG_FindOrMakePointerType(cv_basic_types[T_UQUAD]);
cv_basic_types[T_32PREAL32] = DEBUG_FindOrMakePointerType(cv_basic_types[T_REAL32]);
cv_basic_types[T_32PREAL64] = DEBUG_FindOrMakePointerType(cv_basic_types[T_REAL64]);
cv_basic_types[T_32PRCHAR] = DEBUG_FindOrMakePointerType(cv_basic_types[T_RCHAR]);
cv_basic_types[T_32PWCHAR] = DEBUG_FindOrMakePointerType(cv_basic_types[T_WCHAR]);
cv_basic_types[T_32PINT4] = DEBUG_FindOrMakePointerType(cv_basic_types[T_INT4]);
cv_basic_types[T_32PUINT4] = DEBUG_FindOrMakePointerType(cv_basic_types[T_UINT4]);
}
/*
* In this function, we keep track of deferred debugging information
* that we may need later if we were to need to use the internal debugger.
* We don't fully process it here for performance reasons.
*/
int
DEBUG_RegisterDebugInfo( HMODULE32 hModule, const char *module_name,
u_long v_addr, u_long size)
{
int has_codeview = FALSE;
int rtn = FALSE;
int orig_size;
LPIMAGE_DEBUG_DIRECTORY dbgptr;
struct deferred_debug_info * deefer;
orig_size = size;
dbgptr = (LPIMAGE_DEBUG_DIRECTORY) (hModule + v_addr);
for(; size >= sizeof(*dbgptr); size -= sizeof(*dbgptr), dbgptr++ )
{
switch(dbgptr->Type)
{
case IMAGE_DEBUG_TYPE_CODEVIEW:
case IMAGE_DEBUG_TYPE_MISC:
has_codeview = TRUE;
break;
}
}
size = orig_size;
dbgptr = (LPIMAGE_DEBUG_DIRECTORY) (hModule + v_addr);
for(; size >= sizeof(*dbgptr); size -= sizeof(*dbgptr), dbgptr++ )
{
switch(dbgptr->Type)
{
case IMAGE_DEBUG_TYPE_COFF:
/*
* If we have both codeview and COFF debug info, ignore the
* coff debug info as it would just confuse us, and it is
* less complete.
*
* FIXME - this is broken - if we cannot find the PDB file, then
* we end up with no debugging info at all. In this case, we
* should use the COFF info as a backup.
*/
if( has_codeview )
{
break;
}
case IMAGE_DEBUG_TYPE_CODEVIEW:
case IMAGE_DEBUG_TYPE_MISC:
/*
* This is usually an indirection to a .DBG file.
* This is similar to (but a slightly older format) from the
* PDB file.
*
* First check to see if the image was 'stripped'. If so, it
* means that this entry points to a .DBG file. Otherwise,
* it just points to itself, and we can ignore this.
*/
if( (dbgptr->Type == IMAGE_DEBUG_TYPE_MISC)
&& (PE_HEADER(hModule)->FileHeader.Characteristics & IMAGE_FILE_DEBUG_STRIPPED) == 0 )
{
break;
}
deefer = (struct deferred_debug_info *) xmalloc(sizeof(*deefer));
deefer->module = hModule;
deefer->load_addr = (char *)hModule;
deefer->dbg_info = NULL;
deefer->dbg_size = 0;
/*
* Read the important bits. What we do after this depends
* upon the type, but this is always enough so we are able
* to proceed if we know what we need to do next.
*/
deefer->dbg_size = dbgptr->SizeOfData;
deefer->dbg_info = (char *)(hModule + dbgptr->PointerToRawData);
deefer->dbgdir = dbgptr;
deefer->next = dbglist;
deefer->loaded = FALSE;
deefer->dbg_index = DEBUG_next_index;
deefer->module_name = xstrdup(module_name);
deefer->sectp = PE_SECTIONS(hModule);
deefer->nsect = PE_HEADER(hModule)->FileHeader.NumberOfSections;
dbglist = deefer;
break;
default:
}
}
DEBUG_next_index++;
return (rtn);
}
/*
* ELF modules are also entered into the list - this is so that we
* can make 'info shared' types of displays possible.
*/
int
DEBUG_RegisterELFDebugInfo(int load_addr, u_long size, char * name)
{
struct deferred_debug_info * deefer;
deefer = (struct deferred_debug_info *) xmalloc(sizeof(*deefer));
deefer->module = 0;
/*
* Read the important bits. What we do after this depends
* upon the type, but this is always enough so we are able
* to proceed if we know what we need to do next.
*/
deefer->dbg_size = size;
deefer->dbg_info = (char *) NULL;
deefer->load_addr = (char *) load_addr;
deefer->dbgdir = NULL;
deefer->next = dbglist;
deefer->loaded = TRUE;
deefer->dbg_index = DEBUG_next_index;
deefer->module_name = xstrdup(name);
dbglist = deefer;
DEBUG_next_index++;
return (TRUE);
}
/*
* Process COFF debugging information embedded in a Win32 application.
*
*/
static
int
DEBUG_ProcessCoff(struct deferred_debug_info * deefer)
{
struct CoffAuxSection * aux;
struct CoffDebug * coff;
struct CoffFiles * coff_files = NULL;
struct CoffLinenum * coff_linetab;
char * coff_strtab;
struct CoffSymbol * coff_sym;
struct CoffSymbol * coff_symbol;
struct CoffFiles * curr_file = NULL;
int i;
int j;
int k;
struct CoffLinenum * linepnt;
int linetab_indx;
char namebuff[9];
char * nampnt;
int naux;
DBG_ADDR new_addr;
int nfiles = 0;
int nfiles_alloc = 0;
struct CoffFiles orig_file;
int rtn = FALSE;
char * this_file = NULL;
coff = (struct CoffDebug *) deefer->dbg_info;
coff_symbol = (struct CoffSymbol *) ((unsigned int) coff + coff->SymbolOffset);
coff_linetab = (struct CoffLinenum *) ((unsigned int) coff + coff->LinenumberOffset);
coff_strtab = (char *) ((unsigned int) coff_symbol + 18*coff->N_Sym);
linetab_indx = 0;
for(i=0; i < coff->N_Sym; i++ )
{
/*
* We do this because some compilers (i.e. gcc) incorrectly
* pad the structure up to a 4 byte boundary. The structure
* is really only 18 bytes long, so we have to manually make sure
* we get it right.
*
* FIXME - there must be a way to have autoconf figure out the
* correct compiler option for this. If it is always gcc, that
* makes life simpler, but I don't want to force this.
*/
coff_sym = (struct CoffSymbol *) ((unsigned int) coff_symbol + 18*i);
naux = coff_sym->NumberOfAuxSymbols;
if( coff_sym->StorageClass == IMAGE_SYM_CLASS_FILE )
{
if( nfiles + 1 >= nfiles_alloc )
{
nfiles_alloc += 10;
coff_files = (struct CoffFiles *) realloc( coff_files,
nfiles_alloc * sizeof(struct CoffFiles));
}
curr_file = coff_files + nfiles;
nfiles++;
curr_file->startaddr = 0xffffffff;
curr_file->endaddr = 0;
curr_file->filename = ((char *) coff_sym) + 18;
curr_file->linetab_offset = -1;
curr_file->linecnt = 0;
curr_file->entries = NULL;
curr_file->neps = curr_file->neps_alloc = 0;
#if 0
fprintf(stderr,"New file %s\n", curr_file->filename);
#endif
i += naux;
continue;
}
/*
* This guy marks the size and location of the text section
* for the current file. We need to keep track of this so
* we can figure out what file the different global functions
* go with.
*/
if( (coff_sym->StorageClass == IMAGE_SYM_CLASS_STATIC)
&& (naux != 0)
&& (coff_sym->Type == 0)
&& (coff_sym->SectionNumber == 1) )
{
aux = (struct CoffAuxSection *) ((unsigned int) coff_sym + 18);
if( curr_file->linetab_offset != -1 )
{
#if 0
fprintf(stderr, "Duplicating sect from %s: %x %x %x %d %d\n",
curr_file->filename,
aux->Length,
aux->NumberOfRelocations,
aux->NumberOfLinenumbers,
aux->Number,
aux->Selection);
fprintf(stderr, "More sect %d %x %d %d %d\n",
coff_sym->SectionNumber,
coff_sym->Value,
coff_sym->Type,
coff_sym->StorageClass,
coff_sym->NumberOfAuxSymbols);
#endif
/*
* Save this so we can copy bits from it.
*/
orig_file = *curr_file;
/*
* Duplicate the file entry. We have no way to describe
* multiple text sections in our current way of handling things.
*/
if( nfiles + 1 >= nfiles_alloc )
{
nfiles_alloc += 10;
coff_files = (struct CoffFiles *) realloc( coff_files,
nfiles_alloc * sizeof(struct CoffFiles));
}
curr_file = coff_files + nfiles;
nfiles++;
curr_file->startaddr = 0xffffffff;
curr_file->endaddr = 0;
curr_file->filename = orig_file.filename;
curr_file->linetab_offset = -1;
curr_file->linecnt = 0;
curr_file->entries = NULL;
curr_file->neps = curr_file->neps_alloc = 0;
}
#if 0
else
{
fprintf(stderr, "New text sect from %s: %x %x %x %d %d\n",
curr_file->filename,
aux->Length,
aux->NumberOfRelocations,
aux->NumberOfLinenumbers,
aux->Number,
aux->Selection);
}
#endif
if( curr_file->startaddr > coff_sym->Value )
{
curr_file->startaddr = coff_sym->Value;
}
if( curr_file->startaddr > coff_sym->Value )
{
curr_file->startaddr = coff_sym->Value;
}
if( curr_file->endaddr < coff_sym->Value + aux->Length )
{
curr_file->endaddr = coff_sym->Value + aux->Length;
}
curr_file->linetab_offset = linetab_indx;
curr_file->linecnt = aux->NumberOfLinenumbers;
linetab_indx += aux->NumberOfLinenumbers;
i += naux;
continue;
}
if( (coff_sym->StorageClass == IMAGE_SYM_CLASS_STATIC)
&& (naux == 0)
&& (coff_sym->SectionNumber == 1) )
{
/*
* This is a normal static function when naux == 0.
* Just register it. The current file is the correct
* one in this instance.
*/
if( coff_sym->N.Name.NotLong )
{
memcpy(namebuff, coff_sym->N.ShortName, 8);
namebuff[8] = '\0';
nampnt = &namebuff[0];
}
else
{
nampnt = coff_strtab + coff_sym->N.Name.StrTaboff;
}
if( nampnt[0] == '_' )
{
nampnt++;
}
new_addr.seg = 0;
new_addr.off = (int) (deefer->load_addr + coff_sym->Value);
if( curr_file->neps + 1 >= curr_file->neps_alloc )
{
curr_file->neps_alloc += 10;
curr_file->entries = (struct name_hash **)
realloc( curr_file->entries,
curr_file->neps_alloc * sizeof(struct name_hash *));
}
#if 0
fprintf(stderr,"\tAdding static symbol %s\n", nampnt);
#endif
curr_file->entries[curr_file->neps++] =
DEBUG_AddSymbol( nampnt, &new_addr, this_file, SYM_WIN32 );
i += naux;
continue;
}
if( (coff_sym->StorageClass == IMAGE_SYM_CLASS_EXTERNAL)
&& ISFCN(coff_sym->Type)
&& (coff_sym->SectionNumber > 0) )
{
if( coff_sym->N.Name.NotLong )
{
memcpy(namebuff, coff_sym->N.ShortName, 8);
namebuff[8] = '\0';
nampnt = &namebuff[0];
}
else
{
nampnt = coff_strtab + coff_sym->N.Name.StrTaboff;
}
if( nampnt[0] == '_' )
{
nampnt++;
}
new_addr.seg = 0;
new_addr.off = (int) (deefer->load_addr + coff_sym->Value);
#if 0
fprintf(stderr, "%d: %x %s\n", i, new_addr.off, nampnt);
fprintf(stderr,"\tAdding global symbol %s\n", nampnt);
#endif
/*
* Now we need to figure out which file this guy belongs to.
*/
this_file = NULL;
for(j=0; j < nfiles; j++)
{
if( coff_files[j].startaddr <= coff_sym->Value
&& coff_files[j].endaddr > coff_sym->Value )
{
this_file = coff_files[j].filename;
break;
}
}
if( coff_files[j].neps + 1 >= coff_files[j].neps_alloc )
{
coff_files[j].neps_alloc += 10;
coff_files[j].entries = (struct name_hash **)
realloc( coff_files[j].entries,
coff_files[j].neps_alloc * sizeof(struct name_hash *));
}
coff_files[j].entries[coff_files[j].neps++] =
DEBUG_AddSymbol( nampnt, &new_addr, this_file, SYM_WIN32 );
i += naux;
continue;
}
if( (coff_sym->StorageClass == IMAGE_SYM_CLASS_EXTERNAL)
&& (coff_sym->SectionNumber > 0) )
{
/*
* Similar to above, but for the case of data symbols.
* These aren't treated as entrypoints.
*/
if( coff_sym->N.Name.NotLong )
{
memcpy(namebuff, coff_sym->N.ShortName, 8);
namebuff[8] = '\0';
nampnt = &namebuff[0];
}
else
{
nampnt = coff_strtab + coff_sym->N.Name.StrTaboff;
}
if( nampnt[0] == '_' )
{
nampnt++;
}
new_addr.seg = 0;
new_addr.off = (int) (deefer->load_addr + coff_sym->Value);
#if 0
fprintf(stderr, "%d: %x %s\n", i, new_addr.off, nampnt);
fprintf(stderr,"\tAdding global data symbol %s\n", nampnt);
#endif
/*
* Now we need to figure out which file this guy belongs to.
*/
DEBUG_AddSymbol( nampnt, &new_addr, NULL, SYM_WIN32 );
i += naux;
continue;
}
if( (coff_sym->StorageClass == IMAGE_SYM_CLASS_STATIC)
&& (naux == 0) )
{
/*
* Ignore these. They don't have anything to do with
* reality.
*/
i += naux;
continue;
}
#if 0
fprintf(stderr,"Skipping unknown entry %d %d %d\n", coff_sym->StorageClass,
coff_sym->SectionNumber, naux);
#endif
/*
* For now, skip past the aux entries.
*/
i += naux;
}
/*
* OK, we now should have a list of files, and we should have a list
* of entrypoints. We need to sort the entrypoints so that we are
* able to tie the line numbers with the given functions within the
* file.
*/
if( coff_files != NULL )
{
for(j=0; j < nfiles; j++)
{
if( coff_files[j].entries != NULL )
{
qsort(coff_files[j].entries, coff_files[j].neps,
sizeof(struct name_hash *), DEBUG_cmp_sym);
}
}
/*
* Now pick apart the line number tables, and attach the entries
* to the given functions.
*/
for(j=0; j < nfiles; j++)
{
i = 0;
for(k=0; k < coff_files[j].linecnt; k++)
{
/*
* Another monstrosity caused by the fact that we are using
* a 6 byte structure, and gcc wants to pad structures to 4 byte
* boundaries. Otherwise we could just index into an array.
*/
linepnt = (struct CoffLinenum *)
((unsigned int) coff_linetab +
6*(coff_files[j].linetab_offset + k));
/*
* If we have spilled onto the next entrypoint, then
* bump the counter..
*/
while(TRUE)
{
DEBUG_GetSymbolAddr(coff_files[j].entries[i+1], &new_addr);
if( (i+1 < coff_files[j].neps)
&& ( ((unsigned int) deefer->load_addr + linepnt->VirtualAddr)
>= new_addr.off) )
{
i++;
}
else
{
break;
}
}
/*
* Add the line number. This is always relative to the
* start of the function, so we need to subtract that offset
* first.
*/
DEBUG_GetSymbolAddr(coff_files[j].entries[i], &new_addr);
DEBUG_AddLineNumber(coff_files[j].entries[i],
linepnt->Linenum,
(unsigned int) deefer->load_addr
+ linepnt->VirtualAddr
- new_addr.off);
}
}
}
rtn = TRUE;
if( coff_files != NULL )
{
for(j=0; j < nfiles; j++)
{
if( coff_files[j].entries != NULL )
{
free(coff_files[j].entries);
}
}
free(coff_files);
}
return (rtn);
}
/*
* Process a codeview line number table. Digestify the thing so that
* we can easily reference the thing when we process the rest of
* the information.
*/
static struct codeview_linetab_hdr *
DEBUG_SnarfLinetab(char * linetab,
int size)
{
int file_segcount;
char filename[PATH_MAX];
unsigned int * filetab;
char * fn;
int i;
int k;
struct codeview_linetab_hdr * lt_hdr;
unsigned int * lt_ptr;
int nfile;
int nseg;
union any_size pnt;
union any_size pnt2;
struct startend * start;
int this_seg;
/*
* Now get the important bits.
*/
pnt = (union any_size) linetab;
nfile = *pnt.s++;
nseg = *pnt.s++;
filetab = (unsigned int *) pnt.c;
/*
* Now count up the number of segments in the file.
*/
nseg = 0;
for(i=0; i<nfile; i++)
{
pnt2 = (union any_size) (linetab + filetab[i]);
nseg += *pnt2.s;
}
/*
* Next allocate the header we will be returning.
* There is one header for each segment, so that we can reach in
* and pull bits as required.
*/
lt_hdr = (struct codeview_linetab_hdr *)
xmalloc((nseg + 1) * sizeof(*lt_hdr));
if( lt_hdr == NULL )
{
goto leave;
}
memset(lt_hdr, 0, sizeof(*lt_hdr) * (nseg+1));
/*
* Now fill the header we will be returning, one for each segment.
* Note that this will basically just contain pointers into the existing
* line table, and we do not actually copy any additional information
* or allocate any additional memory.
*/
this_seg = 0;
for(i=0; i<nfile; i++)
{
/*
* Get the pointer into the segment information.
*/
pnt2 = (union any_size) (linetab + filetab[i]);
file_segcount = *pnt2.s;
pnt2.ui++;
lt_ptr = (unsigned int *) pnt2.c;
start = (struct startend *) (lt_ptr + file_segcount);
/*
* Now snarf the filename for all of the segments for this file.
*/
fn = (unsigned char *) (start + file_segcount);
memset(filename, 0, sizeof(filename));
memcpy(filename, fn + 1, *fn);
fn = strdup(filename);
for(k = 0; k < file_segcount; k++, this_seg++)
{
pnt2 = (union any_size) (linetab + lt_ptr[k]);
lt_hdr[this_seg].start = start[k].start;
lt_hdr[this_seg].end = start[k].end;
lt_hdr[this_seg].sourcefile = fn;
lt_hdr[this_seg].segno = *pnt2.s++;
lt_hdr[this_seg].nline = *pnt2.s++;
lt_hdr[this_seg].offtab = pnt2.ui;
lt_hdr[this_seg].linetab = (unsigned short *)
(pnt2.ui + lt_hdr[this_seg].nline);
}
}
leave:
return lt_hdr;
}
static int
DEBUG_SnarfCodeView( struct deferred_debug_info * deefer,
char * cv_data,
int size,
struct codeview_linetab_hdr * linetab)
{
struct name_hash * curr_func = NULL;
struct wine_locals * curr_sym = NULL;
int i;
int j;
int len;
DBG_ADDR new_addr;
int nsect;
union any_size ptr;
IMAGE_SECTION_HEADER * sectp;
union codeview_symbol * sym;
char symname[PATH_MAX];
struct name_hash * thunk_sym = NULL;
ptr = (union any_size) cv_data;
nsect = deefer->nsect;
sectp = deefer->sectp;
/*
* Skip over the first word. Don't really know what it means, but
* it is useless.
*/
ptr.ui++;
/*
* Loop over the different types of records and whenever we
* find something we are interested in, record it and move on.
*/
while( ptr.c - cv_data < size )
{
sym = (union codeview_symbol *) ptr.c;
if( sym->generic.len - sizeof(int) == (ptr.c - cv_data) )
{
/*
* This happens when we have indirect symbols that VC++ 4.2
* sometimes uses when there isn't a line number table.
* We ignore it - we will process and enter all of the
* symbols in the global symbol table anyways, so there
* isn't much point in keeping track of all of this crap.
*/
break;
}
memset(symname, 0, sizeof(symname));
switch(sym->generic.id)
{
case S_GDATA32:
case S_LDATA32:
case S_PUB32:
/*
* First, a couple of sanity checks.
*/
if( sym->data.namelen == 0 )
{
break;
}
if( sym->data.seg == 0 || sym->data.seg > nsect )
{
break;
}
/*
* Global and local data symbols. We don't associate these
* with any given source file.
*/
memcpy(symname, sym->data.name, sym->data.namelen);
new_addr.seg = 0;
new_addr.type = DEBUG_GetCVType(sym->data.symtype);
new_addr.off = (unsigned int) deefer->load_addr +
sectp[sym->data.seg - 1].VirtualAddress +
sym->data.offset;
DEBUG_AddSymbol( symname, &new_addr, NULL, SYM_WIN32 | SYM_DATA );
break;
case S_THUNK32:
/*
* Sort of like a global function, but it just points
* to a thunk, which is a stupid name for what amounts to
* a PLT slot in the normal jargon that everyone else uses.
*/
memcpy(symname, sym->thunk.name, sym->thunk.namelen);
new_addr.seg = 0;
new_addr.type = NULL;
new_addr.off = (unsigned int) deefer->load_addr +
sectp[sym->thunk.segment - 1].VirtualAddress +
sym->thunk.offset;
thunk_sym = DEBUG_AddSymbol( symname, &new_addr, NULL,
SYM_WIN32 | SYM_FUNC);
DEBUG_SetSymbolSize(thunk_sym, sym->thunk.thunk_len);
break;
case S_GPROC32:
case S_LPROC32:
/*
* Global and static functions.
*/
memcpy(symname, sym->proc.name, sym->proc.namelen);
new_addr.seg = 0;
new_addr.type = DEBUG_GetCVType(sym->proc.proctype);
new_addr.off = (unsigned int) deefer->load_addr +
sectp[sym->proc.segment - 1].VirtualAddress +
sym->proc.offset;
/*
* See if we can find a segment that this goes with. If so,
* it means that we also may have line number information
* for this function.
*/
for(i=0; linetab[i].linetab != NULL; i++)
{
if( ((unsigned int) deefer->load_addr
+ sectp[linetab[i].segno - 1].VirtualAddress
+ linetab[i].start <= new_addr.off)
&& ((unsigned int) deefer->load_addr
+ sectp[linetab[i].segno - 1].VirtualAddress
+ linetab[i].end > new_addr.off) )
{
break;
}
}
DEBUG_Normalize(curr_func);
if( linetab[i].linetab == NULL )
{
curr_func = DEBUG_AddSymbol( symname, &new_addr, NULL,
SYM_WIN32 | SYM_FUNC);
}
else
{
/*
* First, create the entry. Then dig through the linetab
* and add whatever line numbers are appropriate for this
* function.
*/
curr_func = DEBUG_AddSymbol( symname, &new_addr,
linetab[i].sourcefile,
SYM_WIN32 | SYM_FUNC);
for(j=0; j < linetab[i].nline; j++)
{
if( linetab[i].offtab[j] >= sym->proc.offset
&& linetab[i].offtab[j] < sym->proc.offset
+ sym->proc.proc_len )
{
DEBUG_AddLineNumber(curr_func, linetab[i].linetab[j],
linetab[i].offtab[j] - sym->proc.offset);
}
}
}
/*
* Add information about where we should set breakpoints
* in this function.
*/
DEBUG_SetSymbolBPOff(curr_func, sym->proc.debug_start);
DEBUG_SetSymbolSize(curr_func, sym->proc.proc_len);
break;
case S_BPREL32:
/*
* Function parameters and stack variables.
*/
memcpy(symname, sym->stack.name, sym->stack.namelen);
curr_sym = DEBUG_AddLocal(curr_func,
0,
sym->stack.offset,
0,
0,
symname);
DEBUG_SetLocalSymbolType(curr_sym, DEBUG_GetCVType(sym->stack.symtype));
break;
default:
break;
}
/*
* Adjust pointer to point to next entry, rounding up to a word
* boundary. MS preserving alignment? Stranger things have
* happened.
*/
if( sym->generic.id == S_PROCREF
|| sym->generic.id == S_DATAREF
|| sym->generic.id == S_UNKNOWN )
{
len = (sym->generic.len + 3) & ~3;
len += ptr.c[16] + 1;
ptr.c += (len + 3) & ~3;
}
else
{
ptr.c += (sym->generic.len + 3) & ~3;
}
}
if( linetab != NULL )
{
free(linetab);
}
return TRUE;
}
/*
* Process PDB file which contains debug information.
*
* These are really weird beasts. They are intended to be incrementally
* updated by the incremental linker, and this means that you need to
* be able to remove and add information. Thus the PDB file is sort of
* like a block structured device, with a freelist and lists of extent numbers
* that are used to get the relevant pieces. In all cases seen so far, the
* blocksize is always 0x400 bytes. The header has a field which apparently
* holds the blocksize, so if it ever changes we are safe.
*
* In general, every time we need to extract something from the pdb file,
* it is easier to copy it into another buffer so we have the information
* in one contiguous block rather than attempt to try and keep track of when
* we need to grab another extent from the pdb file.
*
* The thing that is a real pain about some MS stuff is that they choose
* data structures which are not representable in C. Thus we have to
* hack around and diddle pointers.
*/
/* static */
int
DEBUG_ProcessPDBFile(struct deferred_debug_info * deefer, char * full_filename)
{
char * addr = (char *) 0xffffffff;
unsigned int blocksize;
unsigned int bufflen = 0;
char * buffer = NULL;
unsigned short * extent_table;
int fd = -1;
struct file_ent * fent;
char * filename;
struct file_list * filelist = NULL;
unsigned int gsym_record = 0;
char * gsymtab = NULL;
struct filetab_hdr * hd;
int i;
int j;
unsigned int last_extent;
struct codeview_linetab_hdr * linetab;
unsigned int nblocks;
unsigned int npair;
unsigned int offset;
struct codeview_pdb_hdr * pdbhdr;
unsigned int * pnt;
struct stat statbuf;
int status;
unsigned short * table;
char * toc;
unsigned int toc_blocks;
/*
* FIXME - we should use some kind of search path mechanism to locate
* PDB files. Right now we just look in the current working directory,
* which works much of the time, I guess. Ideally we should be able to
* map the filename back using the settings in wine.ini and perhaps
* we could find it there. This bit of coding is left as an exercise
* for the reader. :-).
*/
filename = strrchr(full_filename, '\\');
if( filename == NULL )
{
filename = full_filename;
}
else
{
filename++;
}
status = stat(filename, &statbuf);
if( status == -1 )
{
fprintf(stderr, "Unable to open .PDB file %s\n", filename);
goto leave;
}
/*
* Now open the file, so that we can mmap() it.
*/
fd = open(filename, O_RDONLY);
if( fd == -1 )
{
fprintf(stderr, "Unable to open .DBG file %s\n", filename);
goto leave;
}
/*
* Now mmap() the file.
*/
addr = mmap(0, statbuf.st_size, PROT_READ,
MAP_PRIVATE, fd, 0);
if( addr == (char *) 0xffffffff )
{
fprintf(stderr, "Unable to mmap .DBG file %s\n", filename);
goto leave;
}
/*
* Now that we have the formalities over and done with, we need
* to find the table of contents for the PDB file.
*/
pdbhdr = (struct codeview_pdb_hdr *) addr;
blocksize = pdbhdr->blocksize;
last_extent = (statbuf.st_size + blocksize - 1) / blocksize;
/*
* The TOC itself isn't always contiguous, so we need to extract a few
* extents from the file to form the TOC.
*/
toc_blocks = (pdbhdr->toc_len + blocksize - 1) / blocksize;
toc = (char *) xmalloc(toc_blocks * blocksize);
table = pdbhdr->toc_ext;
for(i=0; i < toc_blocks; i++)
{
memcpy(toc + blocksize*i, addr + table[i]*blocksize, blocksize);
}
/*
* Next build our own table which will have the size and extent block
* list for each record in the PDB file.
*
* The TOC starts out with the number of files. Then it is followed by
* (npair * 2*sizeof(int)) bytes of information, which are pairs of ints.
* The first one is the size of the record (in bytes), and the second one
* is something else which I haven't figured out yet.
*/
pnt = (unsigned int *) toc;
npair = *pnt++;
extent_table = (unsigned short *) ((unsigned int) toc +
npair * 2 * sizeof(int) + sizeof(int));
/*
* Sanity check.
*/
if( sizeof(int) + 2*sizeof(int)*npair > pdbhdr->toc_len )
{
goto leave;
}
filelist = (struct file_list *) xmalloc(npair * sizeof(*filelist));
if( filelist == NULL )
{
goto leave;
}
memset(filelist, 0, npair * sizeof(*filelist));
nblocks = 0;
for(i=0; i < npair; i++)
{
filelist[i].record_len = pnt[i*2];
filelist[i].nextents = (filelist[i].record_len + blocksize - 1)
/ blocksize;
filelist[i].extent_list = extent_table + nblocks;
nblocks += filelist[i].nextents;
/*
* These get filled in later when we parse one of the records.
*/
filelist[i].linetab_offset = 0;
filelist[i].linetab_len = 0;
}
/*
* OK, now walk through the various records and pick out the bits we
* really want to see. Some of the records are extra special, and
* we need to handle these a little bit differently.
*/
for(i=0; i < npair; i++)
{
if( filelist[i].record_len == 0xffffffff )
{
continue;
}
/*
* Make sure our buffer is large enough to hold the record.
*/
if( bufflen < filelist[i].nextents * blocksize )
{
bufflen = filelist[i].nextents * blocksize;
buffer = (char *) realloc(buffer, bufflen);
}
/*
* Do this just for completeness. It makes debugging easier
* if we have a clean indication of where the record ends.
*/
memset(buffer, 0, filelist[i].nextents * blocksize);
/*
* Next, build the record using the extent list.
*/
for(j=0; j < filelist[i].nextents; j++)
{
memcpy(buffer + j * blocksize,
addr + filelist[i].extent_list[j] * blocksize,
blocksize);
}
pnt = (unsigned int *) buffer;
/*
* OK, now figure out what to do with it.
*/
/*
* Always ignore the first entry. It seems to contain a backup copy
* of the TOC (the last time the file was modified??)
*/
if( i == 0 )
{
continue;
}
/*
* The second entry as a id block. It contains a magic number
* to identify the compiler, plus it also contains the timestamp
* which must match the timestamp in the executable.
*/
if( i == 1 )
{
if( ((*pnt != 19950623) && (*pnt != 19950814))
|| (filelist[i].record_len != 0x24)
|| (pnt[1] != ((struct CodeViewDebug *)(deefer->dbg_info))->cv_timestamp) )
{
goto leave;
}
}
/*
* The third entry contains pointers to the global symbol table,
* plus it also contains additional information about each record
* in the PDB file.
*/
if( i == 3 )
{
hd = (struct filetab_hdr *) buffer;
gsym_record = hd->gsym_file;
gsymtab = (char *) xmalloc( filelist[gsym_record].nextents
* blocksize);
memset(gsymtab, 0, filelist[gsym_record].nextents * blocksize);
for(j=0; j < filelist[gsym_record].nextents; j++)
{
memcpy(gsymtab + j * blocksize,
addr + filelist[gsym_record].extent_list[j] * blocksize,
blocksize);
}
/*
* This record also contains information about where in the
* remaining records we will be able to find the start of the
* line number table. We could locate that bit using heuristics,
* but since we have the info handy, we might as well use it.
*/
offset = sizeof(*hd);
while(1==1)
{
fent = (struct file_ent *) (buffer + offset);
if( offset > hd->ftab_len )
{
break;
}
if( fent->file_number == 0 || fent->file_number >= npair )
{
break;
}
filelist[fent->file_number].linetab_offset =
fent->linetab_offset;
filelist[fent->file_number].linetab_len =
fent->linetab_len;
/*
* Figure out the offset of the next entry.
* There is a fixed part of the record and a variable
* length filename which we must also skip past.
*/
offset += ((unsigned int) &fent->filename - (unsigned int) fent)
+ strlen(fent->filename) + 1;
offset += strlen(buffer+offset) + 1;
offset = (offset + 3) & ~3;
}
}
/*
* Two different magic numbers used as dates.
* These indicate the 'type' table.
*/
if( *pnt == 19950410
|| *pnt == 19951122 )
{
DEBUG_ParseTypeTable(buffer, filelist[i].record_len);
continue;
}
/*
* This is something we really want to look at, since it contains
* real debug info. Anything that doesn't match this can be
* ignored for now.
*/
if( *pnt == 1 )
{
/*
* First, snag the line table, if we have one. This always
* occurs at the end of the record, so we take the linetab
* offset as the end of the normal part of the record.
*/
linetab = NULL;
if( filelist[i].linetab_len != 0 )
{
linetab = DEBUG_SnarfLinetab(buffer + filelist[i].linetab_offset,
filelist[i].linetab_len);
DEBUG_SnarfCodeView(deefer, buffer,
filelist[i].linetab_offset,
linetab);
}
else
{
DEBUG_SnarfCodeView(deefer, buffer,
filelist[i].record_len,
linetab);
}
continue;
}
}
/*
* Finally, process the global symbol table itself. There isn't
* a line number component to this, so we just toss everything
* into the mix and it all should work out.
*/
if( gsym_record != 0 )
{
DEBUG_SnarfCodeView(deefer, gsymtab - sizeof(int),
filelist[gsym_record].record_len,
NULL);
}
leave:
if( gsymtab != NULL )
{
free(gsymtab);
gsymtab = NULL;
}
if( buffer != NULL )
{
free(buffer);
}
if( filelist != NULL )
{
free(filelist);
}
if( addr != (char *) 0xffffffff )
{
munmap(addr, statbuf.st_size);
}
if( fd != -1 )
{
close(fd);
}
return TRUE;
}
/*
* Process DBG file which contains debug information.
*/
/* static */
int
DEBUG_ProcessDBGFile(struct deferred_debug_info * deefer, char * filename)
{
char * addr = (char *) 0xffffffff;
char * codeview;
struct CV4_DirHead * codeview_dir;
struct CV4_DirEnt * codeview_dent;
LPIMAGE_DEBUG_DIRECTORY dbghdr;
struct deferred_debug_info deefer2;
int fd = -1;
int i;
int j;
struct codeview_linetab_hdr * linetab;
int nsect;
LPIMAGE_SEPARATE_DEBUG_HEADER pdbg = NULL;
IMAGE_SECTION_HEADER * sectp;
struct stat statbuf;
int status;
status = stat(filename, &statbuf);
if( status == -1 )
{
fprintf(stderr, "Unable to open .DBG file %s\n", filename);
goto leave;
}
/*
* Now open the file, so that we can mmap() it.
*/
fd = open(filename, O_RDONLY);
if( fd == -1 )
{
fprintf(stderr, "Unable to open .DBG file %s\n", filename);
goto leave;
}
/*
* Now mmap() the file.
*/
addr = mmap(0, statbuf.st_size, PROT_READ,
MAP_PRIVATE, fd, 0);
if( addr == (char *) 0xffffffff )
{
fprintf(stderr, "Unable to mmap .DBG file %s\n", filename);
goto leave;
}
pdbg = (LPIMAGE_SEPARATE_DEBUG_HEADER) addr;
if( pdbg->TimeDateStamp != deefer->dbgdir->TimeDateStamp )
{
fprintf(stderr, "Warning - %s has incorrect internal timestamp\n",
filename);
goto leave;
}
fprintf(stderr, "Processing symbols from %s...\n", filename);
dbghdr = (LPIMAGE_DEBUG_DIRECTORY) ( addr + sizeof(*pdbg)
+ pdbg->NumberOfSections * sizeof(IMAGE_SECTION_HEADER)
+ pdbg->ExportedNamesSize);
sectp = (LPIMAGE_SECTION_HEADER) ((char *) pdbg + sizeof(*pdbg));
nsect = pdbg->NumberOfSections;
for( i=0; i < pdbg->DebugDirectorySize / sizeof(*pdbg); i++, dbghdr++ )
{
switch(dbghdr->Type)
{
case IMAGE_DEBUG_TYPE_COFF:
/*
* Dummy up a deferred debug header to handle the
* COFF stuff embedded within the DBG file.
*/
memset((char *) &deefer2, 0, sizeof(deefer2));
deefer2.dbg_info = (addr + dbghdr->PointerToRawData);
deefer2.dbg_size = dbghdr->SizeOfData;
deefer2.load_addr = deefer->load_addr;
DEBUG_ProcessCoff(&deefer2);
break;
case IMAGE_DEBUG_TYPE_CODEVIEW:
/*
* This is the older format by which codeview stuff is
* stored, known as the 'NB09' format. Newer executables
* and dlls created by VC++ use PDB files instead, which
* have lots of internal similarities, but the overall
* format and structure is quite different.
*/
codeview = (addr + dbghdr->PointerToRawData);
/*
* The first thing in the codeview section should be
* an 'NB09' identifier. As a sanity check, make sure
* it is there.
*/
if( *((unsigned int*) codeview) != 0x3930424e )
{
break;
}
/*
* Next we need to find the directory. This is easy too.
*/
codeview_dir = (struct CV4_DirHead *)
(codeview + ((unsigned int*) codeview)[1]);
/*
* Some more sanity checks. Make sure that everything
* is as we expect it.
*/
if( codeview_dir->next_offset != 0
|| codeview_dir->dhsize != sizeof(*codeview_dir)
|| codeview_dir->desize != sizeof(*codeview_dent) )
{
break;
}
codeview_dent = (struct CV4_DirEnt *) (codeview_dir + 1);
for(j=0; j < codeview_dir->ndir; j++, codeview_dent++)
{
if( codeview_dent->subsect_number == sstAlignSym )
{
/*
* Check the previous entry. If it is a
* sstSrcModule, it contains the line number
* info for this file.
*/
linetab = NULL;
if( codeview_dent[1].module_number == codeview_dent[0].module_number
&& codeview_dent[1].subsect_number == sstSrcModule )
{
linetab = DEBUG_SnarfLinetab(
codeview + codeview_dent[1].offset,
codeview_dent[1].size);
}
if( codeview_dent[-1].module_number == codeview_dent[0].module_number
&& codeview_dent[-1].subsect_number == sstSrcModule )
{
linetab = DEBUG_SnarfLinetab(
codeview + codeview_dent[-1].offset,
codeview_dent[-1].size);
}
/*
* Now process the CV stuff.
*/
DEBUG_SnarfCodeView(deefer,
codeview + codeview_dent->offset,
codeview_dent->size,
linetab);
}
}
break;
default:
break;
}
}
leave:
if( addr != (char *) 0xffffffff )
{
munmap(addr, statbuf.st_size);
}
if( fd != -1 )
{
close(fd);
}
return TRUE;
}
int
DEBUG_ProcessDeferredDebug()
{
struct deferred_debug_info * deefer;
struct CodeViewDebug * cvd;
struct MiscDebug * misc;
char * filename;
int last_proc = -1;
DEBUG_InitCVDataTypes();
for(deefer = dbglist; deefer; deefer = deefer->next)
{
if( deefer->loaded )
{
continue;
}
if( last_proc != deefer->dbg_index )
{
fprintf(stderr, " %s",deefer->module_name);
last_proc = deefer->dbg_index;
}
switch(deefer->dbgdir->Type)
{
case IMAGE_DEBUG_TYPE_COFF:
/*
* Standard COFF debug information that VC++ adds when you
* use /debugtype:both with the linker.
*/
#if 0
fprintf(stderr, "Processing COFF symbols...\n");
#endif
DEBUG_ProcessCoff(deefer);
break;
case IMAGE_DEBUG_TYPE_CODEVIEW:
/*
* This is a pointer to a PDB file of some sort.
*/
cvd = (struct CodeViewDebug *) deefer->dbg_info;
if( strcmp(cvd->cv_nbtype, "NB10") != 0 )
{
/*
* Whatever this is, we don't know how to deal with
* it yet.
*/
break;
}
DEBUG_ProcessPDBFile(deefer, cvd->cv_name);
#if 0
fprintf(stderr, "Processing PDB file %s\n", cvd->cv_name);
#endif
break;
case IMAGE_DEBUG_TYPE_MISC:
/*
* A pointer to a .DBG file of some sort. These files
* can contain either CV4 or COFF information. Open
* the file, and try to do the right thing with it.
*/
misc = (struct MiscDebug *) deefer->dbg_info;
filename = strrchr((char *) &misc->Data, '.');
/*
* Ignore the file if it doesn't have a .DBG extension.
*/
if( (filename == NULL)
|| ( (strcmp(filename, ".dbg") != 0)
&& (strcmp(filename, ".DBG") != 0)) )
{
break;
}
filename = (char *) &misc->Data;
/*
* Do the dirty deed...
*/
DEBUG_ProcessDBGFile(deefer, filename);
break;
default:
/*
* We should never get here...
*/
break;
}
}
return TRUE;
}
/***********************************************************************
* DEBUG_InfoShare
*
* Display shared libarary information.
*/
void DEBUG_InfoShare(void)
{
struct deferred_debug_info * deefer;
fprintf(stderr,"Address\t\tModule\tName\n");
for(deefer = dbglist; deefer; deefer = deefer->next)
{
fprintf(stderr,"0x%8.8x\t(%s)\t%s\n", (unsigned int) deefer->load_addr,
deefer->module ? "Win32" : "ELF", deefer->module_name);
}
}