/* -*- tab-width: 8; c-basic-offset: 4 -*- */ /* * File stabs.c - read stabs information from the wine executable itself. * * Copyright (C) 1996, Eric Youngdale. * 1999, 2000 Eric Pouech */ #include "config.h" #include #include #include #ifdef HAVE_SYS_MMAN_H #include #endif #include #include #include #include #ifndef PATH_MAX #define PATH_MAX _MAX_PATH #endif #include "debugger.h" #if defined(__svr4__) || defined(__sun) #define __ELF__ #endif #ifdef __ELF__ #ifdef HAVE_ELF_H # include #endif #ifdef HAVE_LINK_H # include #endif #endif #ifndef N_UNDF #define N_UNDF 0x00 #endif #ifndef STN_UNDEF # define STN_UNDEF 0 #endif #define N_GSYM 0x20 #define N_FUN 0x24 #define N_STSYM 0x26 #define N_LCSYM 0x28 #define N_MAIN 0x2a #define N_ROSYM 0x2c #define N_OPT 0x3c #define N_RSYM 0x40 #define N_SLINE 0x44 #define N_SO 0x64 #define N_LSYM 0x80 #define N_BINCL 0x82 #define N_SOL 0x84 #define N_PSYM 0xa0 #define N_EINCL 0xa2 #define N_LBRAC 0xc0 #define N_EXCL 0xc2 #define N_RBRAC 0xe0 typedef struct tagELF_DBG_INFO { unsigned long elf_addr; } ELF_DBG_INFO; struct stab_nlist { union { char *n_name; struct stab_nlist *n_next; long n_strx; } n_un; unsigned char n_type; char n_other; short n_desc; unsigned long n_value; }; static void stab_strcpy(char * dest, int sz, const char * source) { /* * A strcpy routine that stops when we hit the ':' character. * Faster than copying the whole thing, and then nuking the * ':'. */ while(*source != '\0' && *source != ':' && sz-- > 0) *dest++ = *source++; *dest = '\0'; assert(sz > 0); } typedef struct { char* name; unsigned long value; int idx; struct datatype** vector; int nrofentries; } include_def; #define MAX_INCLUDES 512 static include_def* include_defs = NULL; static int num_include_def = 0; static int num_alloc_include_def = 0; static int cu_include_stack[MAX_INCLUDES]; static int cu_include_stk_idx = 0; static struct datatype** cu_vector = NULL; static int cu_nrofentries = 0; static int DEBUG_CreateInclude(const char* file, unsigned long val) { if (num_include_def == num_alloc_include_def) { num_alloc_include_def += 256; include_defs = DBG_realloc(include_defs, sizeof(include_defs[0])*num_alloc_include_def); memset(include_defs+num_include_def, 0, sizeof(include_defs[0])*256); } include_defs[num_include_def].name = DBG_strdup(file); include_defs[num_include_def].value = val; include_defs[num_include_def].vector = NULL; include_defs[num_include_def].nrofentries = 0; return num_include_def++; } static int DEBUG_FindInclude(const char* file, unsigned long val) { int i; for (i = 0; i < num_include_def; i++) { if (val == include_defs[i].value && strcmp(file, include_defs[i].name) == 0) return i; } return -1; } static int DEBUG_AddInclude(int idx) { ++cu_include_stk_idx; /* is this happen, just bump MAX_INCLUDES */ /* we could also handle this as another dynarray */ assert(cu_include_stk_idx < MAX_INCLUDES); cu_include_stack[cu_include_stk_idx] = idx; return cu_include_stk_idx; } static void DEBUG_ResetIncludes(void) { /* * The datatypes that we would need to use are reset when * we start a new file. (at least the ones in filenr == 0 */ cu_include_stk_idx = 0;/* keep 0 as index for the .c file itself */ memset(cu_vector, 0, sizeof(cu_vector[0]) * cu_nrofentries); } static void DEBUG_FreeIncludes(void) { int i; DEBUG_ResetIncludes(); for (i = 0; i < num_include_def; i++) { DBG_free(include_defs[i].name); DBG_free(include_defs[i].vector); } DBG_free(include_defs); include_defs = NULL; num_include_def = 0; num_alloc_include_def = 0; DBG_free(cu_vector); cu_vector = NULL; cu_nrofentries = 0; } static struct datatype** DEBUG_FileSubNr2StabEnum(int filenr, int subnr) { struct datatype** ret; /* DEBUG_Printf(DBG_CHN_MESG, "creating type id for (%d,%d)\n", filenr, subnr); */ /* FIXME: I could perhaps create a dummy include_def for each compilation * unit which would allow not to handle those two cases separately */ if (filenr == 0) { if (cu_nrofentries <= subnr) { cu_vector = DBG_realloc(cu_vector, sizeof(cu_vector[0])*(subnr+1)); memset(cu_vector+cu_nrofentries, 0, sizeof(cu_vector[0])*(subnr+1-cu_nrofentries)); cu_nrofentries = subnr + 1; } ret = &cu_vector[subnr]; } else { include_def* idef; assert(filenr <= cu_include_stk_idx); idef = &include_defs[cu_include_stack[filenr]]; if (idef->nrofentries <= subnr) { idef->vector = DBG_realloc(idef->vector, sizeof(idef->vector[0])*(subnr+1)); memset(idef->vector + idef->nrofentries, 0, sizeof(idef->vector[0])*(subnr+1-idef->nrofentries)); idef->nrofentries = subnr + 1; } ret = &idef->vector[subnr]; } /* DEBUG_Printf(DBG_CHN_MESG,"(%d,%d) is %d\n",filenr,subnr,ret); */ return ret; } static struct datatype** DEBUG_ReadTypeEnum(char **x) { int filenr,subnr; if (**x=='(') { (*x)++; /* '(' */ filenr=strtol(*x,x,10); /* */ (*x)++; /* ',' */ subnr=strtol(*x,x,10); /* */ (*x)++; /* ')' */ } else { filenr = 0; subnr = strtol(*x,x,10); /* */ } return DEBUG_FileSubNr2StabEnum(filenr,subnr); } struct ParseTypedefData { char* ptr; char buf[1024]; int idx; }; static int DEBUG_PTS_ReadTypedef(struct ParseTypedefData* ptd, const char* typename, struct datatype** dt); static int DEBUG_PTS_ReadID(struct ParseTypedefData* ptd) { char* first = ptd->ptr; unsigned int len; if ((ptd->ptr = strchr(ptd->ptr, ':')) == NULL) return -1; len = ptd->ptr - first; if (len >= sizeof(ptd->buf) - ptd->idx) return -1; memcpy(ptd->buf + ptd->idx, first, len); ptd->buf[ptd->idx + len] = '\0'; ptd->idx += len + 1; ptd->ptr++; /* ':' */ return 0; } static int DEBUG_PTS_ReadNum(struct ParseTypedefData* ptd, int* v) { char* last; *v = strtol(ptd->ptr, &last, 10); if (last == ptd->ptr) return -1; ptd->ptr = last; return 0; } static int DEBUG_PTS_ReadTypeReference(struct ParseTypedefData* ptd, int* filenr, int* subnr) { if (*ptd->ptr == '(') { /* '(' ',' ')' */ ptd->ptr++; if (DEBUG_PTS_ReadNum(ptd, filenr) == -1) return -1; if (*ptd->ptr++ != ',') return -1; if (DEBUG_PTS_ReadNum(ptd, subnr) == -1) return -1; if (*ptd->ptr++ != ')') return -1; } else { *filenr = 0; if (DEBUG_PTS_ReadNum(ptd, subnr) == -1) return -1; } return 0; } static int DEBUG_PTS_ReadRange(struct ParseTypedefData* ptd, struct datatype** dt, int* lo, int* hi) { /* type ';' ';' ';' */ if (DEBUG_PTS_ReadTypedef(ptd, NULL, dt) == -1) return -1; if (*ptd->ptr++ != ';') return -1; /* ';' */ if (DEBUG_PTS_ReadNum(ptd, lo) == -1) return -1; if (*ptd->ptr++ != ';') return -1; /* ';' */ if (DEBUG_PTS_ReadNum(ptd, hi) == -1) return -1; if (*ptd->ptr++ != ';') return -1; /* ';' */ return 0; } static inline int DEBUG_PTS_ReadAggregate(struct ParseTypedefData* ptd, struct datatype* sdt) { int sz, ofs; char* last; struct datatype* adt; int idx; int doadd; sz = strtol(ptd->ptr, &last, 10); if (last == ptd->ptr) return -1; ptd->ptr = last; doadd = DEBUG_SetStructSize(sdt, sz); /* if the structure has already been filled, just redo the parsing * but don't store results into the struct * FIXME: there's a quite ugly memory leak in there... */ /* Now parse the individual elements of the structure/union. */ while (*ptd->ptr != ';') { /* agg_name : type ',' ',' */ idx = ptd->idx; if (DEBUG_PTS_ReadID(ptd) == -1) return -1; if (DEBUG_PTS_ReadTypedef(ptd, NULL, &adt) == -1) return -1; if (!adt) return -1; if (*ptd->ptr++ != ',') return -1; if (DEBUG_PTS_ReadNum(ptd, &ofs) == -1) return -1; if (*ptd->ptr++ != ',') return -1; if (DEBUG_PTS_ReadNum(ptd, &sz) == -1) return -1; if (*ptd->ptr++ != ';') return -1; if (doadd) DEBUG_AddStructElement(sdt, ptd->buf + idx, adt, ofs, sz); ptd->idx = idx; } ptd->ptr++; /* ';' */ return 0; } static inline int DEBUG_PTS_ReadEnum(struct ParseTypedefData* ptd, struct datatype* edt) { int ofs; int idx; while (*ptd->ptr != ';') { idx = ptd->idx; if (DEBUG_PTS_ReadID(ptd) == -1) return -1; if (DEBUG_PTS_ReadNum(ptd, &ofs) == -1) return -1; if (*ptd->ptr++ != ',') return -1; DEBUG_AddStructElement(edt, ptd->buf + idx, NULL, ofs, 0); ptd->idx = idx; } ptd->ptr++; return 0; } static inline int DEBUG_PTS_ReadArray(struct ParseTypedefData* ptd, struct datatype* adt) { int lo, hi; struct datatype* rdt; /* ar;;; */ if (*ptd->ptr++ != 'r') return -1; /* FIXME: range type is lost, always assume int */ if (DEBUG_PTS_ReadRange(ptd, &rdt, &lo, &hi) == -1) return -1; if (DEBUG_PTS_ReadTypedef(ptd, NULL, &rdt) == -1) return -1; DEBUG_SetArrayParams(adt, lo, hi, rdt); return 0; } static int DEBUG_PTS_ReadTypedef(struct ParseTypedefData* ptd, const char* typename, struct datatype** ret_dt) { int idx, lo, hi; struct datatype* new_dt = NULL; /* newly created data type */ struct datatype* ref_dt; /* referenced data type (pointer...) */ struct datatype* dt1; /* intermediate data type (scope is limited) */ struct datatype* dt2; /* intermediate data type: t1=t2=new_dt */ int filenr1, subnr1; int filenr2 = 0, subnr2 = 0; /* things are a bit complicated because of the way the typedefs are stored inside * the file (we cannot keep the struct datatype** around, because address can * change when realloc is done, so we must call over and over * DEBUG_FileSubNr2StabEnum to keep the correct values around * (however, keeping struct datatype* is valid */ if (DEBUG_PTS_ReadTypeReference(ptd, &filenr1, &subnr1) == -1) return -1; while (*ptd->ptr == '=') { ptd->ptr++; if (new_dt) { DEBUG_Printf(DBG_CHN_MESG, "Bad recursion (1) in typedef\n"); return -1; } /* first handle attribute if any */ switch (*ptd->ptr) { case '@': if (*++ptd->ptr == 's') { ptd->ptr++; if (DEBUG_PTS_ReadNum(ptd, &lo) == -1) { DEBUG_Printf(DBG_CHN_MESG, "Not an attribute... NIY\n"); ptd->ptr -= 2; return -1; } if (*ptd->ptr++ != ';') return -1; } break; } /* then the real definitions */ switch (*ptd->ptr++) { case '*': new_dt = DEBUG_NewDataType(DT_POINTER, NULL); if (DEBUG_PTS_ReadTypedef(ptd, NULL, &ref_dt) == -1) return -1; DEBUG_SetPointerType(new_dt, ref_dt); break; case '(': ptd->ptr--; /* doit a two level by hand, otherwise we'd need a stack */ if (filenr2 || subnr2) { DEBUG_Printf(DBG_CHN_MESG, "Bad recursion (2) in typedef\n"); return -1; } if (DEBUG_PTS_ReadTypeReference(ptd, &filenr2, &subnr2) == -1) return -1; dt1 = *DEBUG_FileSubNr2StabEnum(filenr1, subnr1); dt2 = *DEBUG_FileSubNr2StabEnum(filenr2, subnr2); if (!dt1 && dt2) { new_dt = dt2; filenr2 = subnr2 = 0; } else if (!dt1 && !dt2) { new_dt = NULL; } else { DEBUG_Printf(DBG_CHN_MESG, "Unknown condition %08lx %08lx (%s)\n", (unsigned long)dt1, (unsigned long)dt2, ptd->ptr); return -1; } break; case 'a': new_dt = DEBUG_NewDataType(DT_ARRAY, NULL); if (DEBUG_PTS_ReadArray(ptd, new_dt) == -1) return -1; break; case 'r': new_dt = DEBUG_NewDataType(DT_BASIC, typename); assert(!*DEBUG_FileSubNr2StabEnum(filenr1, subnr1)); *DEBUG_FileSubNr2StabEnum(filenr1, subnr1) = new_dt; if (DEBUG_PTS_ReadRange(ptd, &ref_dt, &lo, &hi) == -1) return -1; /* should perhaps do more here... */ break; case 'f': new_dt = DEBUG_NewDataType(DT_FUNC, NULL); if (DEBUG_PTS_ReadTypedef(ptd, NULL, &ref_dt) == -1) return -1; DEBUG_SetPointerType(new_dt, ref_dt); break; case 'e': new_dt = DEBUG_NewDataType(DT_ENUM, NULL); if (DEBUG_PTS_ReadEnum(ptd, new_dt) == -1) return -1; break; case 's': case 'u': /* dt1 can have been already defined in a forward definition */ dt1 = *DEBUG_FileSubNr2StabEnum(filenr1, subnr1); dt2 = DEBUG_TypeCast(DT_STRUCT, typename); if (!dt1) { new_dt = DEBUG_NewDataType(DT_STRUCT, typename); /* we need to set it here, because a struct can hold a pointer * to itself */ *DEBUG_FileSubNr2StabEnum(filenr1, subnr1) = new_dt; } else { if (DEBUG_GetType(dt1) != DT_STRUCT) { DEBUG_Printf(DBG_CHN_MESG, "Forward declaration is not an aggregate\n"); return -1; } /* should check typename is the same too */ new_dt = dt1; } if (DEBUG_PTS_ReadAggregate(ptd, new_dt) == -1) return -1; break; case 'x': switch (*ptd->ptr++) { case 'e': lo = DT_ENUM; break; case 's': case 'u': lo = DT_STRUCT; break; default: return -1; } idx = ptd->idx; if (DEBUG_PTS_ReadID(ptd) == -1) return -1; new_dt = DEBUG_NewDataType(lo, ptd->buf + idx); ptd->idx = idx; break; default: DEBUG_Printf(DBG_CHN_MESG, "Unknown type '%c'\n", *ptd->ptr); return -1; } } if ((filenr2 || subnr2) && !*DEBUG_FileSubNr2StabEnum(filenr2, subnr2)) { if (!new_dt) { /* this should be a basic type, define it, or even void */ new_dt = DEBUG_NewDataType(DT_BASIC, typename); } *DEBUG_FileSubNr2StabEnum(filenr2, subnr2) = new_dt; } if (!new_dt) { dt1 = *DEBUG_FileSubNr2StabEnum(filenr1, subnr1); if (!dt1) { DEBUG_Printf(DBG_CHN_MESG, "Nothing has been defined <%s>\n", ptd->ptr); return -1; } *ret_dt = dt1; return 0; } *DEBUG_FileSubNr2StabEnum(filenr1, subnr1) = *ret_dt = new_dt; #if 0 if (typename) { DEBUG_Printf(DBG_CHN_MESG, "Adding (%d,%d) %s => ", filenr1, subnr1, typename); DEBUG_PrintTypeCast(new_dt); DEBUG_Printf(DBG_CHN_MESG, "\n"); } #endif return 0; } static int DEBUG_ParseTypedefStab(char* ptr, const char* typename) { struct ParseTypedefData ptd; struct datatype* dt; int ret = -1; /* check for already existing definition */ ptd.idx = 0; if ((ptd.ptr = strchr(ptr, ':'))) { ptd.ptr++; if (*ptd.ptr != '(') ptd.ptr++; ret = DEBUG_PTS_ReadTypedef(&ptd, typename, &dt); } if (ret == -1 || *ptd.ptr) { DEBUG_Printf(DBG_CHN_MESG, "failure on %s at %s\n", ptr, ptd.ptr); return FALSE; } return TRUE; } static struct datatype * DEBUG_ParseStabType(const char * stab) { char * c; /* * Look through the stab definition, and figure out what datatype * this represents. If we have something we know about, assign the * type. */ c = strchr(stab, ':'); if( c == NULL ) return NULL; c++; /* * The next character says more about the type (i.e. data, function, etc) * of symbol. Skip it. */ if (*c != '(') c++; /* * The next is either an integer or a (integer,integer). * The DEBUG_ReadTypeEnum takes care that stab_types is large enough. */ return *DEBUG_ReadTypeEnum(&c); } enum DbgInfoLoad DEBUG_ParseStabs(char * addr, unsigned int load_offset, unsigned int staboff, int stablen, unsigned int strtaboff, int strtablen) { struct name_hash * curr_func = NULL; struct wine_locals * curr_loc = NULL; struct name_hash * curr_sym = NULL; char currpath[PATH_MAX]; int i; int in_external_file = FALSE; int last_nso = -1; unsigned int len; DBG_VALUE new_value; int nstab; char * ptr; char * stabbuff; unsigned int stabbufflen; struct stab_nlist * stab_ptr; char * strs; int strtabinc; char * subpath = NULL; char symname[4096]; nstab = stablen / sizeof(struct stab_nlist); stab_ptr = (struct stab_nlist *) (addr + staboff); strs = (char *) (addr + strtaboff); memset(currpath, 0, sizeof(currpath)); /* * Allocate a buffer into which we can build stab strings for cases * where the stab is continued over multiple lines. */ stabbufflen = 65536; stabbuff = (char *) DBG_alloc(stabbufflen); strtabinc = 0; stabbuff[0] = '\0'; for(i=0; i < nstab; i++, stab_ptr++ ) { ptr = strs + (unsigned int) stab_ptr->n_un.n_name; if( ptr[strlen(ptr) - 1] == '\\' ) { /* * Indicates continuation. Append this to the buffer, and go onto the * next record. Repeat the process until we find a stab without the * '/' character, as this indicates we have the whole thing. */ len = strlen(ptr); if( strlen(stabbuff) + len > stabbufflen ) { stabbufflen += 65536; stabbuff = (char *) DBG_realloc(stabbuff, stabbufflen); } strncat(stabbuff, ptr, len - 1); continue; } else if( stabbuff[0] != '\0' ) { strcat( stabbuff, ptr); ptr = stabbuff; } if( strchr(ptr, '=') != NULL ) { /* * The stabs aren't in writable memory, so copy it over so we are * sure we can scribble on it. */ if( ptr != stabbuff ) { strcpy(stabbuff, ptr); ptr = stabbuff; } stab_strcpy(symname, sizeof(symname), ptr); if (!DEBUG_ParseTypedefStab(ptr, symname)) { /* skip this definition */ stabbuff[0] = '\0'; continue; } } switch(stab_ptr->n_type) { case N_GSYM: /* * These are useless with ELF. They have no value, and you have to * read the normal symbol table to get the address. Thus we * ignore them, and when we process the normal symbol table * we should do the right thing. * * With a.out or mingw, they actually do make some amount of sense. */ new_value.addr.seg = 0; new_value.type = DEBUG_ParseStabType(ptr); new_value.addr.off = load_offset + stab_ptr->n_value; new_value.cookie = DV_TARGET; stab_strcpy(symname, sizeof(symname), ptr); #ifdef __ELF__ curr_sym = DEBUG_AddSymbol( symname, &new_value, currpath, SYM_WINE | SYM_DATA | SYM_INVALID ); #else curr_sym = DEBUG_AddSymbol( symname, &new_value, currpath, SYM_WINE | SYM_DATA ); #endif break; case N_RBRAC: case N_LBRAC: /* * We need to keep track of these so we get symbol scoping * right for local variables. For now, we just ignore them. * The hooks are already there for dealing with this however, * so all we need to do is to keep count of the nesting level, * and find the RBRAC for each matching LBRAC. */ break; case N_LCSYM: case N_STSYM: /* * These are static symbols and BSS symbols. */ new_value.addr.seg = 0; new_value.type = DEBUG_ParseStabType(ptr); new_value.addr.off = load_offset + stab_ptr->n_value; new_value.cookie = DV_TARGET; stab_strcpy(symname, sizeof(symname), ptr); curr_sym = DEBUG_AddSymbol( symname, &new_value, currpath, SYM_WINE | SYM_DATA ); break; case N_PSYM: /* * These are function parameters. */ if( curr_func != NULL && !in_external_file ) { stab_strcpy(symname, sizeof(symname), ptr); curr_loc = DEBUG_AddLocal( curr_func, 0, stab_ptr->n_value, 0, 0, symname ); DEBUG_SetLocalSymbolType( curr_loc, DEBUG_ParseStabType(ptr) ); } break; case N_RSYM: if( curr_func != NULL && !in_external_file ) { stab_strcpy(symname, sizeof(symname), ptr); curr_loc = DEBUG_AddLocal( curr_func, stab_ptr->n_value + 1, 0, 0, 0, symname ); DEBUG_SetLocalSymbolType( curr_loc, DEBUG_ParseStabType(ptr) ); } break; case N_LSYM: if( curr_func != NULL && !in_external_file ) { stab_strcpy(symname, sizeof(symname), ptr); curr_loc = DEBUG_AddLocal( curr_func, 0, stab_ptr->n_value, 0, 0, symname ); DEBUG_SetLocalSymbolType( curr_loc, DEBUG_ParseStabType(ptr) ); } break; case N_SLINE: /* * This is a line number. These are always relative to the start * of the function (N_FUN), and this makes the lookup easier. */ if( curr_func != NULL && !in_external_file ) { #ifdef __ELF__ DEBUG_AddLineNumber(curr_func, stab_ptr->n_desc, stab_ptr->n_value); #else #if 0 /* * This isn't right. The order of the stabs is different under * a.out, and as a result we would end up attaching the line * number to the wrong function. */ DEBUG_AddLineNumber(curr_func, stab_ptr->n_desc, stab_ptr->n_value - curr_func->addr.off); #endif #endif } break; case N_FUN: /* * First, clean up the previous function we were working on. */ DEBUG_Normalize(curr_func); /* * For now, just declare the various functions. Later * on, we will add the line number information and the * local symbols. */ if( !in_external_file) { stab_strcpy(symname, sizeof(symname), ptr); if (*symname) { new_value.addr.seg = 0; new_value.type = DEBUG_ParseStabType(ptr); new_value.addr.off = load_offset + stab_ptr->n_value; new_value.cookie = DV_TARGET; /* * Copy the string to a temp buffer so we * can kill everything after the ':'. We do * it this way because otherwise we end up dirtying * all of the pages related to the stabs, and that * sucks up swap space like crazy. */ #ifdef __ELF__ curr_func = DEBUG_AddSymbol( symname, &new_value, currpath, SYM_WINE | SYM_FUNC | SYM_INVALID ); #else curr_func = DEBUG_AddSymbol( symname, &new_value, currpath, SYM_WINE | SYM_FUNC ); #endif } else { /* some GCC seem to use a N_FUN "" to mark the end of a function */ curr_func = NULL; } } else { /* * Don't add line number information for this function * any more. */ curr_func = NULL; } break; case N_SO: /* * This indicates a new source file. Append the records * together, to build the correct path name. */ #ifndef __ELF__ /* * With a.out, there is no NULL string N_SO entry at the end of * the file. Thus when we find non-consecutive entries, * we consider that a new file is started. */ if( last_nso < i-1 ) { currpath[0] = '\0'; DEBUG_Normalize(curr_func); curr_func = NULL; } #endif if( *ptr == '\0' ) { /* * Nuke old path. */ currpath[0] = '\0'; DEBUG_Normalize(curr_func); curr_func = NULL; } else { if (*ptr != '/') strcat(currpath, ptr); else strcpy(currpath, ptr); subpath = ptr; DEBUG_ResetIncludes(); } last_nso = i; break; case N_SOL: /* * This indicates we are including stuff from an include file. * If this is the main source, enable the debug stuff, otherwise * ignore it. */ in_external_file = !(subpath == NULL || strcmp(ptr, subpath) == 0); break; case N_UNDF: strs += strtabinc; strtabinc = stab_ptr->n_value; DEBUG_Normalize(curr_func); curr_func = NULL; break; case N_OPT: /* * Ignore this. We don't care what it points to. */ break; case N_BINCL: DEBUG_AddInclude(DEBUG_CreateInclude(ptr, stab_ptr->n_value)); break; case N_EINCL: break; case N_EXCL: DEBUG_AddInclude(DEBUG_FindInclude(ptr, stab_ptr->n_value)); break; case N_MAIN: /* * Always ignore these. GCC doesn't even generate them. */ break; default: DEBUG_Printf(DBG_CHN_MESG, "Unknown stab type 0x%02x\n", stab_ptr->n_type); break; } stabbuff[0] = '\0'; #if 0 DEBUG_Printf(DBG_CHN_MESG, "%d %x %s\n", stab_ptr->n_type, (unsigned int) stab_ptr->n_value, strs + (unsigned int) stab_ptr->n_un.n_name); #endif } DEBUG_FreeIncludes(); return DIL_LOADED; } #ifdef __ELF__ /* * Walk through the entire symbol table and add any symbols we find there. * This can be used in cases where we have stripped ELF shared libraries, * or it can be used in cases where we have data symbols for which the address * isn't encoded in the stabs. * * This is all really quite easy, since we don't have to worry about line * numbers or local data variables. */ static int DEBUG_ProcessElfSymtab(DBG_MODULE* module, char* addr, u_long load_addr, Elf32_Shdr* symtab, Elf32_Shdr* strtab) { char * curfile = NULL; struct name_hash * curr_sym = NULL; int flags; int i; DBG_VALUE new_value; int nsym; char * strp; char * symname; Elf32_Sym * symp; symp = (Elf32_Sym *) (addr + symtab->sh_offset); nsym = symtab->sh_size / sizeof(*symp); strp = (char *) (addr + strtab->sh_offset); for(i=0; i < nsym; i++, symp++) { /* * Ignore certain types of entries which really aren't of that much * interest. */ if( ELF32_ST_TYPE(symp->st_info) == STT_SECTION || symp->st_shndx == STN_UNDEF ) { continue; } symname = strp + symp->st_name; /* * Save the name of the current file, so we have a way of tracking * static functions/data. */ if( ELF32_ST_TYPE(symp->st_info) == STT_FILE ) { curfile = symname; continue; } /* * See if we already have something for this symbol. * If so, ignore this entry, because it would have come from the * stabs or from a previous symbol. If the value is different, * we will have to keep the darned thing, because there can be * multiple local symbols by the same name. */ if( (DEBUG_GetSymbolValue(symname, -1, &new_value, FALSE ) == TRUE) && (new_value.addr.off == (load_addr + symp->st_value)) ) continue; new_value.addr.seg = 0; new_value.type = NULL; new_value.addr.off = load_addr + symp->st_value; new_value.cookie = DV_TARGET; flags = SYM_WINE | ((ELF32_ST_TYPE(symp->st_info) == STT_FUNC) ? SYM_FUNC : SYM_DATA); if( ELF32_ST_BIND(symp->st_info) == STB_GLOBAL ) curr_sym = DEBUG_AddSymbol( symname, &new_value, NULL, flags ); else curr_sym = DEBUG_AddSymbol( symname, &new_value, curfile, flags ); /* * Record the size of the symbol. This can come in handy in * some cases. Not really used yet, however. */ if( symp->st_size != 0 ) DEBUG_SetSymbolSize(curr_sym, symp->st_size); } return TRUE; } /* * Loads the symbolic information from ELF module stored in 'filename' * the module has been loaded at 'load_offset' address, so symbols' address * relocation is performed * returns * -1 if the file cannot be found/opened * 0 if the file doesn't contain symbolic info (or this info cannot be * read or parsed) * 1 on success */ enum DbgInfoLoad DEBUG_LoadElfStabs(DBG_MODULE* module) { enum DbgInfoLoad dil = DIL_ERROR; char* addr = (char*)0xffffffff; int fd = -1; struct stat statbuf; Elf32_Ehdr* ehptr; Elf32_Shdr* spnt; char* shstrtab; int i; int stabsect; int stabstrsect; if (module->type != DMT_ELF || ! module->elf_info) { DEBUG_Printf(DBG_CHN_ERR, "Bad elf module '%s'\n", module->module_name); return DIL_ERROR; } /* check that the file exists, and that the module hasn't been loaded yet */ if (stat(module->module_name, &statbuf) == -1) goto leave; if (S_ISDIR(statbuf.st_mode)) goto leave; /* * Now open the file, so that we can mmap() it. */ if ((fd = open(module->module_name, O_RDONLY)) == -1) goto leave; dil = DIL_NOINFO; /* * Now mmap() the file. */ addr = mmap(0, statbuf.st_size, PROT_READ, MAP_PRIVATE, fd, 0); if (addr == (char*)0xffffffff) goto leave; /* * Next, we need to find a few of the internal ELF headers within * this thing. We need the main executable header, and the section * table. */ ehptr = (Elf32_Ehdr*) addr; spnt = (Elf32_Shdr*) (addr + ehptr->e_shoff); shstrtab = (addr + spnt[ehptr->e_shstrndx].sh_offset); stabsect = stabstrsect = -1; for (i = 0; i < ehptr->e_shnum; i++) { if (strcmp(shstrtab + spnt[i].sh_name, ".stab") == 0) stabsect = i; if (strcmp(shstrtab + spnt[i].sh_name, ".stabstr") == 0) stabstrsect = i; } if (stabsect == -1 || stabstrsect == -1) { DEBUG_Printf(DBG_CHN_WARN, "no .stab section\n"); goto leave; } /* * OK, now just parse all of the stabs. */ if (DEBUG_ParseStabs(addr, module->elf_info->elf_addr, spnt[stabsect].sh_offset, spnt[stabsect].sh_size, spnt[stabstrsect].sh_offset, spnt[stabstrsect].sh_size)) { dil = DIL_LOADED; } else { dil = DIL_ERROR; DEBUG_Printf(DBG_CHN_WARN, "bad stabs\n"); goto leave; } for (i = 0; i < ehptr->e_shnum; i++) { if ( (strcmp(shstrtab + spnt[i].sh_name, ".symtab") == 0) && (spnt[i].sh_type == SHT_SYMTAB)) DEBUG_ProcessElfSymtab(module, addr, module->elf_info->elf_addr, spnt + i, spnt + spnt[i].sh_link); if ( (strcmp(shstrtab + spnt[i].sh_name, ".dynsym") == 0) && (spnt[i].sh_type == SHT_DYNSYM)) DEBUG_ProcessElfSymtab(module, addr, module->elf_info->elf_addr, spnt + i, spnt + spnt[i].sh_link); } leave: if (addr != (char*)0xffffffff) munmap(addr, statbuf.st_size); if (fd != -1) close(fd); return dil; } /* * Loads the information for ELF module stored in 'filename' * the module has been loaded at 'load_offset' address * returns * -1 if the file cannot be found/opened * 0 if the file doesn't contain symbolic info (or this info cannot be * read or parsed) * 1 on success */ static enum DbgInfoLoad DEBUG_ProcessElfFile(const char* filename, unsigned int load_offset, unsigned int* dyn_addr) { enum DbgInfoLoad dil = DIL_ERROR; char* addr = (char*)0xffffffff; int fd = -1; struct stat statbuf; Elf32_Ehdr* ehptr; Elf32_Shdr* spnt; Elf32_Phdr* ppnt; char * shstrtab; int i; DBG_MODULE* module = NULL; DWORD size; DWORD delta; DEBUG_Printf(DBG_CHN_TRACE, "Processing elf file '%s'\n", filename); /* check that the file exists, and that the module hasn't been loaded yet */ if (stat(filename, &statbuf) == -1) goto leave; /* * Now open the file, so that we can mmap() it. */ if ((fd = open(filename, O_RDONLY)) == -1) goto leave; /* * Now mmap() the file. */ addr = mmap(0, statbuf.st_size, PROT_READ, MAP_PRIVATE, fd, 0); if (addr == (char*)0xffffffff) goto leave; dil = DIL_NOINFO; /* * Next, we need to find a few of the internal ELF headers within * this thing. We need the main executable header, and the section * table. */ ehptr = (Elf32_Ehdr*) addr; spnt = (Elf32_Shdr*) (addr + ehptr->e_shoff); shstrtab = (addr + spnt[ehptr->e_shstrndx].sh_offset); /* if non relocatable ELF, then remove fixed address from computation * otherwise, all addresses are zero based */ delta = (load_offset == 0) ? ehptr->e_entry : 0; /* grab size of module once loaded in memory */ ppnt = (Elf32_Phdr*) (addr + ehptr->e_phoff); size = 0; for (i = 0; i < ehptr->e_phnum; i++) { if (ppnt[i].p_type != PT_LOAD) continue; if (size < ppnt[i].p_vaddr - delta + ppnt[i].p_memsz) size = ppnt[i].p_vaddr - delta + ppnt[i].p_memsz; } for (i = 0; i < ehptr->e_shnum; i++) { if (strcmp(shstrtab + spnt[i].sh_name, ".bss") == 0 && spnt[i].sh_type == SHT_NOBITS) { if (size < spnt[i].sh_addr - delta + spnt[i].sh_size) size = spnt[i].sh_addr - delta + spnt[i].sh_size; } if (strcmp(shstrtab + spnt[i].sh_name, ".dynamic") == 0 && spnt[i].sh_type == SHT_DYNAMIC) { if (dyn_addr) *dyn_addr = spnt[i].sh_addr; } } module = DEBUG_RegisterELFModule((load_offset == 0) ? ehptr->e_entry : load_offset, size, filename); if (!module) { dil = DIL_ERROR; goto leave; } if ((module->elf_info = DBG_alloc(sizeof(ELF_DBG_INFO))) == NULL) { DEBUG_Printf(DBG_CHN_ERR, "OOM\n"); exit(0); } module->elf_info->elf_addr = load_offset; dil = DEBUG_LoadElfStabs(module); leave: if (addr != (char*)0xffffffff) munmap(addr, statbuf.st_size); if (fd != -1) close(fd); if (module) module->dil = dil; return dil; } static enum DbgInfoLoad DEBUG_ProcessElfFileFromPath(const char * filename, unsigned int load_offset, unsigned int* dyn_addr, const char* path) { enum DbgInfoLoad dil = DIL_ERROR; char *s, *t, *fn; char* paths = NULL; if (!path) return -1; for (s = paths = DBG_strdup(path); s && *s; s = (t) ? (t+1) : NULL) { t = strchr(s, ':'); if (t) *t = '\0'; fn = (char*)DBG_alloc(strlen(filename) + 1 + strlen(s) + 1); if (!fn) break; strcpy(fn, s ); strcat(fn, "/"); strcat(fn, filename); dil = DEBUG_ProcessElfFile(fn, load_offset, dyn_addr); DBG_free(fn); if (dil != DIL_ERROR) break; s = (t) ? (t+1) : NULL; } DBG_free(paths); return dil; } static enum DbgInfoLoad DEBUG_ProcessElfObject(const char* filename, unsigned int load_offset, unsigned int* dyn_addr) { enum DbgInfoLoad dil = DIL_ERROR; if (filename == NULL) return DIL_ERROR; if (DEBUG_FindModuleByName(filename, DMT_ELF)) return DIL_LOADED; dil = DEBUG_ProcessElfFile(filename, load_offset, dyn_addr); /* if relative pathname, try some absolute base dirs */ if (dil == DIL_ERROR && !strchr(filename, '/')) { dil = DEBUG_ProcessElfFileFromPath(filename, load_offset, dyn_addr, getenv("PATH")); if (dil == DIL_ERROR) dil = DEBUG_ProcessElfFileFromPath(filename, load_offset, dyn_addr, getenv("LD_LIBRARY_PATH")); } DEBUG_ReportDIL(dil, "ELF", filename, load_offset); return dil; } static BOOL DEBUG_WalkList(struct r_debug* dbg_hdr) { u_long lm_addr; struct link_map lm; Elf32_Ehdr ehdr; char bufstr[256]; /* * Now walk the linked list. In all known ELF implementations, * the dynamic loader maintains this linked list for us. In some * cases the first entry doesn't appear with a name, in other cases it * does. */ for (lm_addr = (u_long)dbg_hdr->r_map; lm_addr; lm_addr = (u_long)lm.l_next) { if (!DEBUG_READ_MEM_VERBOSE((void*)lm_addr, &lm, sizeof(lm))) return FALSE; if (lm.l_addr != 0 && DEBUG_READ_MEM_VERBOSE((void*)lm.l_addr, &ehdr, sizeof(ehdr)) && ehdr.e_type == ET_DYN && /* only look at dynamic modules */ lm.l_name != NULL && DEBUG_READ_MEM_VERBOSE((void*)lm.l_name, bufstr, sizeof(bufstr))) { bufstr[sizeof(bufstr) - 1] = '\0'; DEBUG_ProcessElfObject(bufstr, (unsigned)lm.l_addr, NULL); } } return TRUE; } static BOOL DEBUG_RescanElf(void) { struct r_debug dbg_hdr; if (!DEBUG_CurrProcess || !DEBUG_READ_MEM_VERBOSE((void*)DEBUG_CurrProcess->dbg_hdr_addr, &dbg_hdr, sizeof(dbg_hdr))) return FALSE; switch (dbg_hdr.r_state) { case RT_CONSISTENT: DEBUG_WalkList(&dbg_hdr); break; case RT_ADD: break; case RT_DELETE: /* FIXME: this is not currently handled, would need some kind of mark&sweep algo */ break; } return FALSE; } enum DbgInfoLoad DEBUG_ReadExecutableDbgInfo(const char* exe_name) { Elf32_Dyn dyn; struct r_debug dbg_hdr; enum DbgInfoLoad dil = DIL_NOINFO; unsigned int dyn_addr; /* * Make sure we can stat and open this file. */ if (exe_name == NULL) goto leave; DEBUG_ProcessElfObject(exe_name, 0, &dyn_addr); do { if (!DEBUG_READ_MEM_VERBOSE((void*)dyn_addr, &dyn, sizeof(dyn))) goto leave; dyn_addr += sizeof(dyn); } while (dyn.d_tag != DT_DEBUG && dyn.d_tag != DT_NULL); if (dyn.d_tag == DT_NULL) goto leave; /* * OK, now dig into the actual tables themselves. */ if (!DEBUG_READ_MEM_VERBOSE((void*)dyn.d_un.d_ptr, &dbg_hdr, sizeof(dbg_hdr))) goto leave; assert(!DEBUG_CurrProcess->dbg_hdr_addr); DEBUG_CurrProcess->dbg_hdr_addr = (u_long)dyn.d_un.d_ptr; if (dbg_hdr.r_brk) { DBG_VALUE value; DEBUG_Printf(DBG_CHN_TRACE, "Setting up a breakpoint on r_brk(%lx)\n", (unsigned long)dbg_hdr.r_brk); DEBUG_SetBreakpoints(FALSE); value.type = NULL; value.cookie = DV_TARGET; value.addr.seg = 0; value.addr.off = (DWORD)dbg_hdr.r_brk; DEBUG_AddBreakpoint(&value, DEBUG_RescanElf); DEBUG_SetBreakpoints(TRUE); } dil = DEBUG_WalkList(&dbg_hdr); leave: return dil; } #else /* !__ELF__ */ enum DbgInfoLoad DEBUG_ReadExecutableDbgInfo(const char* exe_name) { return FALSE; } #endif /* __ELF__ */