void destroy_lib_info(void) {
while (debuggee.libs) {
lib_info *next = debuggee.libs->next;
if (debuggee.libs->symtab) destroy_symtab(debuggee.libs->symtab);
free(debuggee.libs);
debuggee.libs = next;
}
}
void destroy_lib_info(struct ps_prochandle* ph) {
lib_info* lib = ph->libs;
while (lib) {
lib_info* next = lib->next;
if (lib->symtab) {
destroy_symtab(lib->symtab);
}
free(lib);
lib = next;
}
}
int
main ( int argc, char **argv )
{
yyparse();
simplify_tree ( &root, root );
// node_print(root, 0);
find_globals();
size_t n_globals = tlhash_size(global_names);
symbol_t *global_list[n_globals];
tlhash_values ( global_names, (void **)&global_list );
for ( size_t i=0; i<n_globals; i++ )
if ( global_list[i]->type == SYM_FUNCTION )
bind_names ( global_list[i], global_list[i]->node );
// print_globals();
generate_program ();
destroy_subtree ( root );
destroy_symtab();
}
// read symbol table from given fd.
struct symtab* build_symtab(int fd) {
symtab_t* symtab = NULL;
int i;
mach_header_64 header;
off_t image_start;
if (!get_arch_off(fd, CPU_TYPE_X86_64, &image_start)) {
print_debug("failed in get fat header\n");
return NULL;
}
lseek(fd, image_start, SEEK_SET);
if (read(fd, (void *)&header, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
print_debug("reading header failed!\n");
return NULL;
}
// header
if (header.magic != MH_MAGIC_64) {
print_debug("not a valid .dylib file\n");
return NULL;
}
load_command lcmd;
symtab_command symtabcmd;
nlist_64 lentry;
bool lcsymtab_exist = false;
long filepos = ltell(fd);
for (i = 0; i < header.ncmds; i++) {
lseek(fd, filepos, SEEK_SET);
if (read(fd, (void *)&lcmd, sizeof(load_command)) != sizeof(load_command)) {
print_debug("read load_command failed for file\n");
return NULL;
}
filepos += lcmd.cmdsize; // next command position
if (lcmd.cmd == LC_SYMTAB) {
lseek(fd, -sizeof(load_command), SEEK_CUR);
lcsymtab_exist = true;
break;
}
}
if (!lcsymtab_exist) {
print_debug("No symtab command found!\n");
return NULL;
}
if (read(fd, (void *)&symtabcmd, sizeof(symtab_command)) != sizeof(symtab_command)) {
print_debug("read symtab_command failed for file");
return NULL;
}
symtab = (symtab_t *)malloc(sizeof(symtab_t));
if (symtab == NULL) {
print_debug("out of memory: allocating symtab\n");
return NULL;
}
// create hash table, we use berkeley db to
// manipulate the hash table.
symtab->hash_table = dbopen(NULL, O_CREAT | O_RDWR, 0600, DB_HASH, NULL);
if (symtab->hash_table == NULL)
goto quit;
symtab->num_symbols = symtabcmd.nsyms;
symtab->symbols = (symtab_symbol *)malloc(sizeof(symtab_symbol) * symtab->num_symbols);
symtab->strs = (char *)malloc(sizeof(char) * symtabcmd.strsize);
if (symtab->symbols == NULL || symtab->strs == NULL) {
print_debug("out of memory: allocating symtab.symbol or symtab.strs\n");
goto quit;
}
lseek(fd, image_start + symtabcmd.symoff, SEEK_SET);
for (i = 0; i < symtab->num_symbols; i++) {
if (read(fd, (void *)&lentry, sizeof(nlist_64)) != sizeof(nlist_64)) {
print_debug("read nlist_64 failed at %i\n", i);
goto quit;
}
symtab->symbols[i].offset = lentry.n_value;
symtab->symbols[i].size = lentry.n_un.n_strx; // index
}
// string table
lseek(fd, image_start + symtabcmd.stroff, SEEK_SET);
int size = read(fd, (void *)(symtab->strs), symtabcmd.strsize * sizeof(char));
if (size != symtabcmd.strsize * sizeof(char)) {
print_debug("reading string table failed\n");
goto quit;
}
for (i = 0; i < symtab->num_symbols; i++) {
symtab->symbols[i].name = symtab->strs + symtab->symbols[i].size;
if (i > 0) {
// fix size
symtab->symbols[i - 1].size = symtab->symbols[i].size - symtab->symbols[i - 1].size;
print_debug("%s size = %d\n", symtab->symbols[i - 1].name, symtab->symbols[i - 1].size);
}
if (i == symtab->num_symbols - 1) {
// last index
symtab->symbols[i].size =
symtabcmd.strsize - symtab->symbols[i].size;
print_debug("%s size = %d\n", symtab->symbols[i].name, symtab->symbols[i].size);
}
}
// build a hashtable for fast query
build_search_table(symtab);
return symtab;
quit:
if (symtab) destroy_symtab(symtab);
return NULL;
}
// read symbol table from given fd.
struct symtab* build_symtab(int fd) {
ELF_EHDR ehdr;
struct symtab* symtab = NULL;
// Reading of elf header
struct elf_section *scn_cache = NULL;
int cnt = 0;
ELF_SHDR* shbuf = NULL;
ELF_SHDR* cursct = NULL;
ELF_PHDR* phbuf = NULL;
int symtab_found = 0;
int dynsym_found = 0;
uint32_t symsection = SHT_SYMTAB;
uintptr_t baseaddr = (uintptr_t)-1;
lseek(fd, (off_t)0L, SEEK_SET);
if (! read_elf_header(fd, &ehdr)) {
// not an elf
return NULL;
}
// read ELF header
if ((shbuf = read_section_header_table(fd, &ehdr)) == NULL) {
goto quit;
}
baseaddr = find_base_address(fd, &ehdr);
scn_cache = calloc(ehdr.e_shnum, sizeof(*scn_cache));
if (scn_cache == NULL) {
goto quit;
}
for (cursct = shbuf, cnt = 0; cnt < ehdr.e_shnum; cnt++) {
scn_cache[cnt].c_shdr = cursct;
if (cursct->sh_type == SHT_SYMTAB ||
cursct->sh_type == SHT_STRTAB ||
cursct->sh_type == SHT_DYNSYM) {
if ( (scn_cache[cnt].c_data = read_section_data(fd, &ehdr, cursct)) == NULL) {
goto quit;
}
}
if (cursct->sh_type == SHT_SYMTAB)
symtab_found++;
if (cursct->sh_type == SHT_DYNSYM)
dynsym_found++;
cursct++;
}
if (!symtab_found && dynsym_found)
symsection = SHT_DYNSYM;
for (cnt = 1; cnt < ehdr.e_shnum; cnt++) {
ELF_SHDR *shdr = scn_cache[cnt].c_shdr;
if (shdr->sh_type == symsection) {
ELF_SYM *syms;
int j, n;
size_t size;
// FIXME: there could be multiple data buffers associated with the
// same ELF section. Here we can handle only one buffer. See man page
// for elf_getdata on Solaris.
// guarantee(symtab == NULL, "multiple symtab");
symtab = calloc(1, sizeof(*symtab));
if (symtab == NULL) {
goto quit;
}
// the symbol table
syms = (ELF_SYM *)scn_cache[cnt].c_data;
// number of symbols
n = shdr->sh_size / shdr->sh_entsize;
// create hash table, we use berkeley db to
// manipulate the hash table.
symtab->hash_table = dbopen(NULL, O_CREAT | O_RDWR, 0600, DB_HASH, NULL);
// guarantee(symtab->hash_table, "unexpected failure: dbopen");
if (symtab->hash_table == NULL)
goto bad;
// shdr->sh_link points to the section that contains the actual strings
// for symbol names. the st_name field in ELF_SYM is just the
// string table index. we make a copy of the string table so the
// strings will not be destroyed by elf_end.
size = scn_cache[shdr->sh_link].c_shdr->sh_size;
symtab->strs = malloc(size);
if (symtab->strs == NULL)
goto bad;
memcpy(symtab->strs, scn_cache[shdr->sh_link].c_data, size);
// allocate memory for storing symbol offset and size;
symtab->num_symbols = n;
symtab->symbols = calloc(n , sizeof(*symtab->symbols));
if (symtab->symbols == NULL)
goto bad;
// copy symbols info our symtab and enter them info the hash table
for (j = 0; j < n; j++, syms++) {
DBT key, value;
char *sym_name = symtab->strs + syms->st_name;
// skip non-object and non-function symbols
int st_type = ELF_ST_TYPE(syms->st_info);
if ( st_type != STT_FUNC && st_type != STT_OBJECT)
continue;
// skip empty strings and undefined symbols
if (*sym_name == '\0' || syms->st_shndx == SHN_UNDEF) continue;
symtab->symbols[j].name = sym_name;
symtab->symbols[j].offset = syms->st_value - baseaddr;
symtab->symbols[j].size = syms->st_size;
key.data = sym_name;
key.size = strlen(sym_name) + 1;
value.data = &(symtab->symbols[j]);
value.size = sizeof(symtab_symbol);
(*symtab->hash_table->put)(symtab->hash_table, &key, &value, 0);
}
}
}
goto quit;
bad:
destroy_symtab(symtab);
symtab = NULL;
quit:
if (shbuf) free(shbuf);
if (phbuf) free(phbuf);
if (scn_cache) {
for (cnt = 0; cnt < ehdr.e_shnum; cnt++) {
if (scn_cache[cnt].c_data != NULL) {
free(scn_cache[cnt].c_data);
}
}
free(scn_cache);
}
return symtab;
}
