930 lines
29 KiB
C++
930 lines
29 KiB
C++
#include <mach-o/dyld.h>
|
|
#include <mach/mach.h>
|
|
#include <stdint.h>
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
|
|
char *pwd;
|
|
uint32_t pwd_len;
|
|
|
|
int custom_strcmp(const char *p1, const char *p2) {
|
|
const unsigned char *s1 = (const unsigned char *)p1;
|
|
const unsigned char *s2 = (const unsigned char *)p2;
|
|
unsigned char c1, c2;
|
|
do {
|
|
c1 = (unsigned char)*s1++;
|
|
c2 = (unsigned char)*s2++;
|
|
if (c1 == '\0')
|
|
return c1 - c2;
|
|
} while (c1 == c2);
|
|
return c1 - c2;
|
|
}
|
|
|
|
int custom_strncmp(const char *s1, const char *s2, register size_t n)
|
|
{
|
|
register unsigned char u1, u2;
|
|
|
|
while (n-- > 0)
|
|
{
|
|
u1 = (unsigned char) *s1++;
|
|
u2 = (unsigned char) *s2++;
|
|
if (u1 != u2)
|
|
return u1 - u2;
|
|
if (u1 == '\0')
|
|
return 0;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
void set_cwd(const char *const *envp) {
|
|
while (*envp) {
|
|
// PWD=
|
|
if (0x3d445750 == *(uint32_t *)(*envp)) {
|
|
break;
|
|
}
|
|
envp++;
|
|
}
|
|
pwd = (char *)(*envp + 4);
|
|
for (; pwd[pwd_len] != 0; pwd_len++)
|
|
;
|
|
}
|
|
|
|
const uint32_t magic64 = 0xfeedfacf;
|
|
const uint32_t magic32 = 0xfeedface;
|
|
|
|
struct libcache_item {
|
|
void *header;
|
|
void *trie;
|
|
uint32_t trie_size;
|
|
uint32_t hash;
|
|
|
|
uint64_t slide;
|
|
|
|
// pointer to segment address
|
|
uint32_t nsegment;
|
|
uint64_t* segment;
|
|
};
|
|
|
|
struct libcache {
|
|
struct libcache_item *libs;
|
|
uint32_t size;
|
|
|
|
void *main;
|
|
void *thislib;
|
|
void *libdyld;
|
|
};
|
|
|
|
// try these hashes
|
|
// https://gist.github.com/sgsfak/9ba382a0049f6ee885f68621ae86079b
|
|
uint32_t fnv_hash(const char *str) {
|
|
unsigned char *s = (unsigned char *)str; /* unsigned string */
|
|
|
|
/* See the FNV parameters at www.isthe.com/chongo/tech/comp/fnv/#FNV-param */
|
|
const uint32_t FNV_32_PRIME = 0x01000193; /* 16777619 */
|
|
|
|
uint32_t h = 0x811c9dc5; /* 2166136261 */
|
|
while (*s != 0) {
|
|
/* xor the bottom with the current octet */
|
|
h ^= *s++;
|
|
/* multiply by the 32 bit FNV magic prime mod 2^32 */
|
|
h *= FNV_32_PRIME;
|
|
}
|
|
|
|
return h;
|
|
}
|
|
|
|
// calculate the hash to search
|
|
// _dyld_get_image_name returns the full path to the library
|
|
// while the static path in LC_DYLIB (and such) could be relative
|
|
// we should expand the path to fullpath to correctly compute the hash
|
|
//
|
|
// the hardest part is the @rpath, because there can be many LC_RPATH
|
|
// and @rpath can also reference @loader_path
|
|
uint32_t calculate_libname_hash(const libcache *cache, const char *name) {
|
|
uint32_t hash;
|
|
uint32_t (*hash_func)(const char *) = fnv_hash;
|
|
if (name[0] == '.') {
|
|
// resolve relative path with ./ ../ ../../ and so on
|
|
char *p = realpath(name, 0);
|
|
hash = hash_func(p);
|
|
free(p);
|
|
} else if (name[0] == '@') {
|
|
// TODO: resolve @rpath
|
|
// ohyeah this is gonna be wild
|
|
// loop through all rpath and resolve that rpath
|
|
// then resolve the full path for all rpath
|
|
//
|
|
// which rpath is correct can be done by checking if the cache has that hash
|
|
printf("resolver for @rpath is not supported yet\n");
|
|
} else {
|
|
hash = hash_func(name);
|
|
}
|
|
return hash;
|
|
}
|
|
|
|
// dummy no sus function to look for dyld header
|
|
// i don't know if dyld_stub_binder should be better
|
|
// because if they are not familiar with dyld
|
|
// they would not suspect dyld_stub_binder inside modern macho
|
|
// Added iOS 6, macOS 10.8
|
|
extern "C" uint32_t dyld_get_sdk_version(const mach_header *mh);
|
|
void exported_from_c();
|
|
|
|
void decode_uleb128(char *&addr, uint32_t *ret) {
|
|
uint32_t result = 0;
|
|
int shift = 0;
|
|
|
|
while (1) {
|
|
unsigned char byte = *(unsigned char *)(addr);
|
|
addr++;
|
|
|
|
result |= (byte & 0x7f) << shift;
|
|
shift += 7;
|
|
|
|
if (!(byte & 0x80))
|
|
break;
|
|
}
|
|
|
|
*ret = result;
|
|
}
|
|
|
|
void *find_header(void *_func) {
|
|
// Approach 1: (not stable)
|
|
// we assume that text section is small enough to fit on 1 page
|
|
// so the header should stay at the top of the page due to allocation logic
|
|
// the slice/slide is random but always align 0x1000 so we test a few values
|
|
// to see if the magic value is found
|
|
//
|
|
// Guaranteed to stop, but search range is small
|
|
|
|
// const uint64_t page_size = 0x4000;
|
|
// uint64_t func = (uint64_t)_func;
|
|
// uint64_t potential_head = func + (0x4000 - (func % page_size));
|
|
// void* head = 0;
|
|
// for (uint64_t i = 0x1000; i < 0xf000; i+=0x1000) {
|
|
// uint32_t* x = (uint32_t*)(potential_head - i);
|
|
// if (*x == magic64 || *x == magic32) {
|
|
// head = (void*)x;
|
|
// break;
|
|
// }
|
|
// }
|
|
// return head;
|
|
|
|
// Approach 2: (more stable)
|
|
// We know that the header is 0x1000 aligned,
|
|
// just loop until the magic value is found
|
|
// Using while loop so ¯\_(ツ)_/¯
|
|
const uint64_t page_size = 0x1000;
|
|
uint64_t func = (uint64_t)_func;
|
|
uint64_t potential_head = func + (0x1000 - (func % page_size));
|
|
|
|
void *head = 0;
|
|
uint32_t *x = (uint32_t *)(potential_head);
|
|
while (*x != magic64 && *x != magic32) {
|
|
x -= 0x1000 / 4;
|
|
}
|
|
return (void *)x;
|
|
}
|
|
|
|
uint64_t get_slide(const void *header) {
|
|
const uint32_t magic = *(uint32_t *)header;
|
|
char *ptr = (char *)header;
|
|
if (magic == magic64) {
|
|
ptr += 0x20;
|
|
} else {
|
|
ptr += 0x20 - 0x4;
|
|
}
|
|
|
|
uint64_t slice = 0;
|
|
const uint32_t ncmds = *((uint32_t *)header + 4);
|
|
for (int i = 0; i < ncmds; i++) {
|
|
const uint32_t cmd = *((uint32_t *)ptr + 0);
|
|
const uint32_t cmdsize = *((uint32_t *)ptr + 1);
|
|
if (cmd == LC_SEGMENT_64) {
|
|
char *name = (char *)((uint64_t *)ptr + 1);
|
|
uint64_t vmaddr = *((uint64_t *)ptr + 3);
|
|
uint64_t fileoffset = *((uint64_t *)ptr + 5);
|
|
if (custom_strcmp(name, "__TEXT") == 0) {
|
|
slice = (uint64_t)header - vmaddr;
|
|
return slice;
|
|
}
|
|
}
|
|
ptr += cmdsize;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
void print_macho_summary(const void *header) {
|
|
const uint32_t magic = *(uint32_t *)header;
|
|
char *ptr = (char *)header;
|
|
if (magic == magic64) {
|
|
ptr += 0x20;
|
|
} else {
|
|
ptr += 0x20 - 0x4;
|
|
}
|
|
|
|
const uint32_t ncmds = *((uint32_t *)header + 4);
|
|
uint64_t linkedit_vmaddr;
|
|
uint64_t linkedit_fileoffset;
|
|
uint64_t slide;
|
|
printf("parsing macho at %p\n", header);
|
|
printf("ncmds %x\n", ncmds);
|
|
for (int i = 0; i < ncmds; i++) {
|
|
const uint32_t cmd = *((uint32_t *)ptr + 0);
|
|
const uint32_t cmdsize = *((uint32_t *)ptr + 1);
|
|
printf(" cmd %x %x\n", cmd, cmdsize);
|
|
if (cmd == LC_DYLD_EXPORTS_TRIE) {
|
|
const uint32_t offset = *((uint32_t *)ptr + 2);
|
|
const uint32_t size = *((uint32_t *)ptr + 3);
|
|
printf(" export trie: offset=0x%x size=0x%x\n", offset, size);
|
|
}
|
|
if (cmd == LC_SEGMENT_64) {
|
|
char *name = (char *)((uint64_t *)ptr + 1);
|
|
uint64_t vmaddr = *((uint64_t *)ptr + 3);
|
|
uint64_t vmsize = *((uint64_t *)ptr + 4);
|
|
uint64_t fileoffset = *((uint64_t *)ptr + 5);
|
|
uint64_t filesize = *((uint64_t *)ptr + 6);
|
|
if (custom_strcmp(name, "__TEXT") == 0) {
|
|
slide = (uint64_t)header - vmaddr;
|
|
printf(" --- slide=0x%llx ---\n", slide);
|
|
} else if (custom_strcmp(name, "__LINKEDIT") == 0) {
|
|
linkedit_vmaddr = vmaddr;
|
|
linkedit_fileoffset = fileoffset;
|
|
}
|
|
printf(" Segment %s\n", name);
|
|
printf(" vmaddr=0x%llx fileoffset=0x%llx\n", vmaddr, fileoffset);
|
|
printf(" vmsize=0x%llx filesize=0x%llx\n", vmsize, filesize);
|
|
|
|
uint64_t nsect = *((uint32_t *)ptr + 8*2);
|
|
char* sections_ptr = (char*)((uint32_t*)ptr + 18);
|
|
for (int sec = 0; sec < nsect; sec++) {
|
|
char* secname = sections_ptr;
|
|
uint64_t addr = *((uint64_t*)sections_ptr + 4);
|
|
uint64_t size = *((uint64_t*)sections_ptr + 5);
|
|
uint32_t fileoffset = *((uint32_t*)sections_ptr + 6*2);
|
|
printf(" Section %s\n", sections_ptr);
|
|
printf(" addr=0x%llx size=0x%llx fileoffset=0x%x\n", addr, size, fileoffset);
|
|
}
|
|
}
|
|
if (cmd == LC_SYMTAB) {
|
|
uint32_t symoff = *((uint32_t *)ptr + 2);
|
|
uint32_t nsym = *((uint32_t *)ptr + 3);
|
|
uint32_t stroff = (*((uint32_t *)ptr + 4));
|
|
uint32_t strsize = *((uint32_t *)ptr + 5);
|
|
|
|
struct symbol_t {
|
|
uint32_t strx;
|
|
uint8_t flags;
|
|
uint8_t sect;
|
|
uint16_t desc;
|
|
uint64_t value;
|
|
};
|
|
|
|
uint64_t symtab_start = (uint64_t)symoff - linkedit_fileoffset + slide + linkedit_vmaddr;
|
|
uint64_t stroff_start = (uint64_t)stroff - linkedit_fileoffset + slide + linkedit_vmaddr;
|
|
|
|
printf(" symtab: offset=0x%x nsym=0x%x\n", symoff, nsym);
|
|
for (int j = 0; j < nsym; j++) {
|
|
struct symbol_t* symtab = (struct symbol_t*)symtab_start;
|
|
struct symbol_t symbol = symtab[j];
|
|
char* name = (char*)stroff_start + symbol.strx;
|
|
printf(" %s %llx => %p\n", name, symbol.value, (void*)(symbol.value + slide));
|
|
}
|
|
}
|
|
if (cmd == LC_REEXPORT_DYLIB) {
|
|
uint32_t name_offset = *((uint32_t *)ptr + 2);
|
|
char *name = (char *)ptr + name_offset;
|
|
printf(" reexport lib %s\n", name);
|
|
}
|
|
ptr += cmdsize;
|
|
}
|
|
}
|
|
|
|
void *get_export_trie(const void *header, uint32_t &size) {
|
|
const uint32_t magic = *(uint32_t *)header;
|
|
char *ptr = (char *)header;
|
|
if (magic == magic64) {
|
|
ptr += 0x20;
|
|
} else {
|
|
ptr += 0x20 - 0x4;
|
|
}
|
|
|
|
uint64_t slice = 0;
|
|
uint64_t linkedit_vmaddr = 0;
|
|
uint64_t linkedit_fileoffset = 0;
|
|
const uint32_t ncmds = *((uint32_t *)header + 4);
|
|
for (int i = 0; i < ncmds; i++) {
|
|
const uint32_t cmd = *((uint32_t *)ptr + 0);
|
|
const uint32_t cmdsize = *((uint32_t *)ptr + 1);
|
|
if (cmd == LC_DYLD_EXPORTS_TRIE) {
|
|
const uint32_t offset = *((uint32_t *)ptr + 2);
|
|
size = *((uint32_t *)ptr + 3);
|
|
uint64_t offset_in_linkedit = (uint64_t)offset - linkedit_fileoffset;
|
|
return (void *)(linkedit_vmaddr + slice + offset_in_linkedit);
|
|
}
|
|
if (cmd == LC_DYLD_INFO_ONLY) {
|
|
const uint32_t offset = *((uint32_t *)ptr + 10);
|
|
size = *((uint32_t *)ptr + 11);
|
|
uint64_t offset_in_linkedit = (uint64_t)offset - linkedit_fileoffset;
|
|
return (void *)(linkedit_vmaddr + slice + offset_in_linkedit);
|
|
}
|
|
if (cmd == LC_SEGMENT_64) {
|
|
char *name = (char *)((uint64_t *)ptr + 1);
|
|
uint64_t vmaddr = *((uint64_t *)ptr + 3);
|
|
uint64_t fileoffset = *((uint64_t *)ptr + 5);
|
|
if (custom_strcmp(name, "__TEXT") == 0) {
|
|
slice = (uint64_t)header - vmaddr;
|
|
} else if (custom_strcmp(name, "__LINKEDIT") == 0) {
|
|
linkedit_vmaddr = vmaddr;
|
|
linkedit_fileoffset = fileoffset;
|
|
}
|
|
}
|
|
ptr += cmdsize;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
uint32_t should_follow_symbol(char *&buffer, char *&_find) {
|
|
// printf("follow check %s has prefix: %s\n", _find, buffer);
|
|
char *find = _find;
|
|
char is_prefix = true;
|
|
while (1) {
|
|
int find_end = *find == 0;
|
|
int buffer_end = *buffer == 0;
|
|
int check = *buffer == *find;
|
|
// printf("check is %x == %x\n", *buffer, *find);
|
|
|
|
if (buffer_end) {
|
|
// we must always run to the end of buffer, marked 0x00
|
|
buffer++;
|
|
break;
|
|
}
|
|
if (find_end) {
|
|
// symbol to find is shorter than current buffer string
|
|
// but we still need to run to the end of buffer
|
|
// so just set not prefix
|
|
is_prefix = false;
|
|
}
|
|
if (!check) {
|
|
is_prefix = false;
|
|
}
|
|
buffer++;
|
|
find++;
|
|
}
|
|
// only move forward if is_prefix
|
|
if (is_prefix) {
|
|
_find = find;
|
|
// printf("prefix is found\n");
|
|
}
|
|
return is_prefix;
|
|
}
|
|
|
|
void *find_in_export_trie(const void *header, void *trie, const char *symbol) {
|
|
uint32_t func = 0;
|
|
|
|
char *ptr = (char *)trie;
|
|
char *find = (char *)symbol;
|
|
while (1) {
|
|
// terminal node will have data
|
|
uint32_t data_count = 0;
|
|
decode_uleb128(ptr, &data_count);
|
|
if (data_count != 0 && *find == 0) {
|
|
// printf("reached terminal node\n");
|
|
break;
|
|
} else if (data_count) {
|
|
// still need to follow the branch
|
|
ptr += data_count;
|
|
}
|
|
char num_child = ptr[0];
|
|
ptr++;
|
|
|
|
int still_following = 0;
|
|
for (char i = 0; i < num_child; i++) {
|
|
still_following = should_follow_symbol(ptr, find);
|
|
uint32_t follow_offset;
|
|
decode_uleb128(ptr, &follow_offset);
|
|
if (still_following) {
|
|
ptr = (char *)trie + follow_offset;
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (!still_following) {
|
|
// symbol not found
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
char count = *(ptr - 1);
|
|
ptr++; // flags
|
|
// uleb128 offset
|
|
decode_uleb128(ptr, &func);
|
|
return (void *)((char *)header + func);
|
|
}
|
|
|
|
void *find_in_lib(struct libcache *cache, struct libcache_item *lib,
|
|
const char *symbol);
|
|
|
|
void *find_in_reexport(struct libcache *cache, struct libcache_item *lib,
|
|
const char *symbol) {
|
|
void *header = lib->header;
|
|
const uint32_t magic = *(uint32_t *)header;
|
|
char *ptr = (char *)header;
|
|
if (magic == magic64) {
|
|
ptr += 0x20;
|
|
} else {
|
|
ptr += 0x20 - 0x4;
|
|
}
|
|
|
|
const uint32_t ncmds = *((uint32_t *)header + 4);
|
|
for (int i = 0; i < ncmds; i++) {
|
|
const uint32_t cmd = *((uint32_t *)ptr + 0);
|
|
const uint32_t cmdsize = *((uint32_t *)ptr + 1);
|
|
if (cmd != LC_REEXPORT_DYLIB) {
|
|
ptr += cmdsize;
|
|
continue;
|
|
}
|
|
uint32_t name_offset = *((uint32_t *)ptr + 2);
|
|
char *name = (char *)ptr + name_offset;
|
|
uint32_t hash = calculate_libname_hash(cache, name);
|
|
for (int j = 0; j < cache->size; j++) {
|
|
struct libcache_item reexport = cache->libs[j];
|
|
if (reexport.hash != hash) {
|
|
continue;
|
|
}
|
|
void *found = find_in_lib(cache, &reexport, symbol);
|
|
if (found)
|
|
return found;
|
|
}
|
|
ptr += cmdsize;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
void *find_in_lib(struct libcache *cache, struct libcache_item *lib,
|
|
const char *symbol) {
|
|
void *direct = find_in_export_trie(lib->header, lib->trie, symbol);
|
|
if (direct)
|
|
return direct;
|
|
// cannot find in directly exported trie, loop through all reexport libs
|
|
return find_in_reexport(cache, lib, symbol);
|
|
}
|
|
|
|
void *custom_dlsym(struct libcache *cache, const char *libname,
|
|
const char *symbol) {
|
|
uint32_t hash = calculate_libname_hash(cache, libname);
|
|
for (int i = 0; i < cache->size; i++) {
|
|
struct libcache_item cache_lib = cache->libs[i];
|
|
if (cache_lib.hash == hash) {
|
|
return find_in_lib(cache, &cache_lib, symbol);
|
|
}
|
|
}
|
|
printf("cannot find lib with hash 0x%x\n", hash);
|
|
return 0;
|
|
}
|
|
|
|
void *custom_dlsym(struct libcache *cache, uint32_t hash, const char *symbol) {
|
|
for (int i = 0; i < cache->size; i++) {
|
|
struct libcache_item cache_lib = cache->libs[i];
|
|
if (cache_lib.hash == hash) {
|
|
return find_in_lib(cache, &cache_lib, symbol);
|
|
}
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
void bootstrap_libcache_item(struct libcache_item* item, const void* header, const char* name) {
|
|
item->header = (void*)header;
|
|
item->trie = get_export_trie(header, item->trie_size);
|
|
|
|
const uint32_t magic = *(uint32_t *)header;
|
|
char *ptr = (char *)header;
|
|
if (magic == magic64) {
|
|
ptr += 0x20;
|
|
} else {
|
|
ptr += 0x20 - 0x4;
|
|
}
|
|
|
|
const uint32_t ncmds = *((uint32_t *)header + 4);
|
|
char* command_ptr = ptr;
|
|
|
|
for (int i = 0; i < ncmds; i++) {
|
|
const uint32_t cmd = *((uint32_t *)ptr + 0);
|
|
const uint32_t cmdsize = *((uint32_t *)ptr + 1);
|
|
if (cmd == LC_SEGMENT_64) {
|
|
char *name = (char *)((uint64_t *)ptr + 1);
|
|
if (custom_strcmp(name, "__TEXT") == 0) {
|
|
uint64_t vmaddr = *((uint64_t *)ptr + 3);
|
|
item->slide = (uint64_t)header - vmaddr;
|
|
}
|
|
item->nsegment++;
|
|
}
|
|
ptr += cmdsize;
|
|
}
|
|
|
|
ptr = command_ptr;
|
|
item->segment = (uint64_t*)malloc(sizeof(uint64_t) * item->nsegment);
|
|
for (int i = 0, segment_i = 0; i < ncmds; i++) {
|
|
const uint32_t cmd = *((uint32_t *)ptr + 0);
|
|
const uint32_t cmdsize = *((uint32_t *)ptr + 1);
|
|
if (cmd == LC_SEGMENT_64) {
|
|
uint64_t vmaddr = *((uint64_t *)ptr + 3);
|
|
item->segment[segment_i++] = (vmaddr + item->slide);
|
|
}
|
|
ptr += cmdsize;
|
|
}
|
|
return;
|
|
|
|
}
|
|
|
|
struct libcache_item* get_libcache_with_name(struct libcache* cache, const char* name) {
|
|
void* to_find = 0;
|
|
if (custom_strcmp(name, "main") == 0) {
|
|
to_find = cache->main;
|
|
} else if (custom_strcmp(name, "thislib") == 0) {
|
|
to_find = cache->thislib;
|
|
}
|
|
uint32_t hash = calculate_libname_hash(cache, name);
|
|
for (int i = 0; i < cache->size; i++) {
|
|
struct libcache_item* cache_lib = &cache->libs[i];
|
|
// search by hash or by pointer for special case
|
|
if (cache_lib->hash == hash || cache_lib->header == to_find) {
|
|
return cache_lib;
|
|
}
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
void dump_export_trie(const void* trie, uint32_t size, const char* filename) {
|
|
FILE *outfile = fopen(filename, "wb");
|
|
fwrite((char*)trie, size, 1, outfile);
|
|
fclose(outfile);
|
|
}
|
|
|
|
void dump_export_trie_of(const char* libname, const libcache* cache, const char* filename) {
|
|
uint32_t hash = calculate_libname_hash(cache, libname);
|
|
for (int i = 0; i < cache->size; i++) {
|
|
struct libcache_item cache_lib = cache->libs[i];
|
|
if (cache_lib.hash == hash) {
|
|
return dump_export_trie(cache_lib.trie, cache_lib.trie_size, filename);
|
|
}
|
|
}
|
|
}
|
|
|
|
void* find_in_symtab(const libcache_item* lib, const char* find) {
|
|
void* header = lib->header;
|
|
const uint32_t magic = *(uint32_t *)header;
|
|
char *ptr = (char *)header;
|
|
if (magic == magic64) {
|
|
ptr += 0x20;
|
|
} else {
|
|
ptr += 0x20 - 0x4;
|
|
}
|
|
|
|
const uint32_t ncmds = *((uint32_t *)header + 4);
|
|
char* command_ptr = ptr;
|
|
|
|
uint64_t linkedit_vmaddr;
|
|
uint64_t linkedit_fileoffset;
|
|
uint64_t slide;
|
|
for (int i = 0; i < ncmds; i++) {
|
|
const uint32_t cmd = *((uint32_t *)ptr + 0);
|
|
const uint32_t cmdsize = *((uint32_t *)ptr + 1);
|
|
if (cmd == LC_SYMTAB) {
|
|
uint32_t symoff = *((uint32_t *)ptr + 2);
|
|
uint32_t nsym = *((uint32_t *)ptr + 3);
|
|
uint32_t stroff = (*((uint32_t *)ptr + 4));
|
|
uint32_t strsize = *((uint32_t *)ptr + 5);
|
|
|
|
struct symbol_t {
|
|
uint32_t strx;
|
|
uint8_t flags;
|
|
uint8_t sect;
|
|
uint16_t desc;
|
|
uint64_t value;
|
|
};
|
|
|
|
uint64_t symtab_start = (uint64_t)symoff - linkedit_fileoffset + slide + linkedit_vmaddr;
|
|
uint64_t stroff_start = (uint64_t)stroff - linkedit_fileoffset + slide + linkedit_vmaddr;
|
|
|
|
for (int j = 0; j < nsym; j++) {
|
|
struct symbol_t* symtab = (struct symbol_t*)symtab_start;
|
|
struct symbol_t symbol = symtab[j];
|
|
char* name = (char*)stroff_start + symbol.strx;
|
|
if (custom_strcmp(name, find) == 0) {
|
|
return (void*)(symbol.value + slide);
|
|
}
|
|
}
|
|
break;
|
|
}
|
|
if (cmd == LC_SEGMENT_64) {
|
|
char *name = (char *)((uint64_t *)ptr + 1);
|
|
uint64_t vmaddr = *((uint64_t *)ptr + 3);
|
|
uint64_t fileoffset = *((uint64_t *)ptr + 5);
|
|
if (custom_strcmp(name, "__TEXT") == 0) {
|
|
slide = (uint64_t)header - vmaddr;
|
|
} else if (custom_strcmp(name, "__LINKEDIT") == 0) {
|
|
linkedit_vmaddr = vmaddr;
|
|
linkedit_fileoffset = fileoffset;
|
|
}
|
|
}
|
|
ptr += cmdsize;
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
void* find_in_symtab(const char* libname, const libcache* cache, const char* find) {
|
|
uint32_t hash = calculate_libname_hash(cache, libname);
|
|
struct libcache_item *cache_lib = 0;
|
|
for (int i = 0; i < cache->size; i++) {
|
|
if (cache->libs[i].hash == hash) {
|
|
cache_lib = &(cache->libs[i]);
|
|
break;
|
|
}
|
|
}
|
|
return find_in_symtab(cache_lib, find);
|
|
}
|
|
int hook_printf(const char *format, ...) {
|
|
va_list args;
|
|
va_start(args, format);
|
|
|
|
printf("HOOKED BEGIN LOL\n");
|
|
int status = printf(format, args);
|
|
printf("HOOKED END LOL\n");
|
|
|
|
va_end(args);
|
|
return status;
|
|
}
|
|
|
|
struct ProgramVars {
|
|
void *mh; // mach_header or mach_header64
|
|
int *NXArgcPtr;
|
|
const char ***NXArgvPtr;
|
|
const char ***environPtr;
|
|
const char **__prognamePtr;
|
|
};
|
|
|
|
void build_cache(struct libcache& cache, void* main);
|
|
void fix(struct libcache& cache);
|
|
|
|
void test(struct libcache& cache);
|
|
|
|
__attribute__((constructor)) static void
|
|
bruh(int argc, const char *const argv[], const char *const envp[],
|
|
const char *const apple[], const struct ProgramVars *vars) {
|
|
printf("=== manual symbol bind starts ===\n");
|
|
set_cwd(envp);
|
|
|
|
// ProgramVars contains pointer to main executable (mapped) file
|
|
|
|
struct libcache cache;
|
|
build_cache(cache, (void *)(vars->mh));
|
|
|
|
// dump_export_trie_of(
|
|
// "/usr/lib/libobjc.A.dylib", &cache,
|
|
// "../scripts/lib_objc_export_trie.bin");
|
|
|
|
// dump_macho(
|
|
// "/usr/lib/libobjc.A.dylib", &cache,
|
|
// "../scripts/lib_objc_symtab.bin");
|
|
|
|
// struct libcache_item* objc = get_libcache_with_name(&cache, "/usr/lib/libobjc.A.dylib");
|
|
// print_macho_summary(objc->header);
|
|
|
|
// test(cache);
|
|
|
|
// ATTENTION:
|
|
// If we choose to resolve **this** lib
|
|
// Before resolve is complete, **DO NOT** call any library function
|
|
//
|
|
// The following functions can be used:
|
|
// (we do not remove them for **our lib**)
|
|
// - malloc
|
|
// - free
|
|
fix(cache);
|
|
|
|
for (int i = 0; i < cache.size; i++) {
|
|
free(cache.libs[i].segment);
|
|
}
|
|
free(cache.libs);
|
|
printf("=== manual symbol bind completes ===\n");
|
|
}
|
|
|
|
void build_cache(struct libcache& cache, void* main) {
|
|
const uint64_t main_slide = get_slide(main);
|
|
// Find our lib (mapped) file
|
|
const void *thislib = find_header((void *)bruh);
|
|
// Find dyld lib (mapped) file using a no-sus function
|
|
const void *libdyld = find_header((void *)dyld_get_sdk_version);
|
|
|
|
cache.main = (void*)main;
|
|
cache.thislib = (void*)thislib;
|
|
cache.libdyld = (void*)libdyld;
|
|
uint32_t libsystem_hash =
|
|
calculate_libname_hash(&cache, "/usr/lib/libSystem.B.dylib");
|
|
|
|
// From libdyld header, we can list exports table
|
|
// to find all function we want to use
|
|
//
|
|
// This way there is no leakage of functions we use to do our trick
|
|
// mostly to hide
|
|
// - _dyld_image_count
|
|
// - _dyld_get_image_name
|
|
// - _dyld_get_image_header
|
|
// - _dyld_get_image_vmaddr_slide
|
|
|
|
// The above functions are crucial to find all libraries loaded
|
|
// From which we will traverse the exports table to replace
|
|
// _got and _la_symbol_pointer data
|
|
|
|
// Our lib can hide more details too
|
|
// We can resolve all functions we use
|
|
// before resolving the main executable imports
|
|
//
|
|
// This will make our lib use only dyld_get_sdk_version
|
|
// For the main executable, imports are empty due to manual resolve
|
|
|
|
printf("executable header at %p\n", main);
|
|
printf("lib header at %p\n", thislib);
|
|
printf("libdyld header at %p\n", libdyld);
|
|
|
|
uint32_t trie_size;
|
|
void *libdyld_export_trie = get_export_trie(libdyld, trie_size);
|
|
|
|
// we have to traverse the trie to find these symbols
|
|
// because if we self-rebuild import table for **this** lib,
|
|
// these symbols aren't resolved
|
|
// so we have to resolve ourselves and then rebuild the symbols for others
|
|
typedef int (*dyld_image_count_t)(void);
|
|
typedef char *(*dyld_get_image_name_t)(int);
|
|
typedef void *(*dyld_get_image_header_t)(int);
|
|
int (*dyld_image_count_func)(void) = (dyld_image_count_t)find_in_export_trie(
|
|
libdyld, libdyld_export_trie, "__dyld_image_count");
|
|
void *(*dyld_get_image_header_func)(int) =
|
|
(dyld_get_image_header_t)find_in_export_trie(libdyld, libdyld_export_trie,
|
|
"__dyld_get_image_header");
|
|
char *(*dyld_get_image_name_func)(int) =
|
|
(dyld_get_image_name_t)find_in_export_trie(libdyld, libdyld_export_trie,
|
|
"__dyld_get_image_name");
|
|
|
|
cache.size = dyld_image_count_func();
|
|
cache.libs =
|
|
(struct libcache_item *)malloc(sizeof(struct libcache_item) * cache.size);
|
|
for (int i = 0; i < cache.size; i++) {
|
|
void *header = dyld_get_image_header_func(i);
|
|
char *name = dyld_get_image_name_func(i);
|
|
bootstrap_libcache_item(&cache.libs[i], header, name);
|
|
cache.libs[i].hash = calculate_libname_hash(&cache, name);
|
|
// printf("%p %s\n", header, name);
|
|
}
|
|
}
|
|
|
|
void fix(struct libcache& cache) {
|
|
// now we have function to find exported symbols
|
|
// it supports full name search or hash search
|
|
// to reserve space, we use the hash search
|
|
//
|
|
// so we will collect all imported symbols, and its offset to fix
|
|
// with legacy symbol resolve
|
|
// __got always has dyld_stub_binder
|
|
// __la_symbol_ptr
|
|
// with modern symbol resolve
|
|
// __got now contains full rebase/bind opcode
|
|
//
|
|
// the list of all imported symbols should be
|
|
// [(offset, name, libhash)]
|
|
// if we want to also fix framework/libraries used by the main executable,
|
|
// (only those that are not governed by the system)
|
|
// we should also have extra list(s) for that lib to resolve ourselves
|
|
//
|
|
// main: [(offset, name, libhash)]
|
|
// libA: [(offset, name, libhash)]
|
|
// libB: [(offset, name, libhash)]
|
|
//
|
|
// using the list is temporary for PoC
|
|
// we know that many symbols are exported from 1 lib
|
|
// so we can build a trie (yes, more trie)
|
|
// where the symbols are now concatenated with libhash 4 bytes as prefix
|
|
// and the offset is at the terminal node
|
|
//
|
|
// this way, we can reduce the libhash, although we need to build a trie
|
|
// build the trie is harder than traversing it
|
|
//
|
|
// just an idea, if we can somehow reduce the datasize then it would be better
|
|
|
|
// NOTES:
|
|
// mach_task_self() has a conflicting symbol or something,
|
|
// in symbol table it's: _mach_task_self_
|
|
// but have to search with: _mach_task_self
|
|
//
|
|
// so future replacement into mach_task_self has to use _mach_task_self
|
|
// despite the symbol is _mach_task_self_
|
|
//
|
|
// may need to look into why this happens so we can deal with this more
|
|
// generic
|
|
|
|
uint32_t libsystem_hash =
|
|
calculate_libname_hash(&cache, "/usr/lib/libSystem.B.dylib");
|
|
|
|
// this is just for now,
|
|
// in the future we will self fix our lib so our lib also has no import
|
|
typedef void *(*vm_protect_t)(void *, uint64_t, uint64_t, int, int);
|
|
typedef void *(*mach_task_self_t)();
|
|
mach_task_self_t mach_task_self_func =
|
|
(mach_task_self_t)custom_dlsym(&cache, libsystem_hash, "_mach_task_self");
|
|
vm_protect_t vm_protect_func =
|
|
(vm_protect_t)custom_dlsym(&cache, libsystem_hash, "_vm_protect");
|
|
|
|
int npage_rw_fixed = 0;
|
|
uint64_t page_rw_fixed[10]; // should be dynamic, but works for now
|
|
|
|
#include "out/b.h"
|
|
// think of a way to get what binary to fix
|
|
// so we can iterate through them
|
|
if (nimports == 0) {
|
|
printf("there is no imports to fix\n");
|
|
}
|
|
struct libcache_item* libfixing = get_libcache_with_name(&cache, lib_to_resolve);
|
|
// print_macho_summary(libfixing->header);
|
|
for (int i = 0; i < nimports; i++) {
|
|
struct imported_symbol symbol = imported_table[i];
|
|
uint64_t fix_at = symbol.offset + libfixing->segment[symbol.segment_i];
|
|
|
|
int need_rw_fix = true;
|
|
for (int j = 0; j < npage_rw_fixed; j++) {
|
|
if (page_rw_fixed[j] <= fix_at &&
|
|
page_rw_fixed[j] + 0x1000 > fix_at) {
|
|
need_rw_fix = false;
|
|
}
|
|
}
|
|
if (need_rw_fix) {
|
|
uint64_t start_page = fix_at - (fix_at % 0x1000);
|
|
vm_protect_func(mach_task_self_func(), start_page, 0x1000, 0,
|
|
VM_PROT_READ | VM_PROT_WRITE);
|
|
page_rw_fixed[npage_rw_fixed++] = start_page;
|
|
printf("modify page starts at 0x%llx to RW\n", start_page);
|
|
}
|
|
|
|
void *resolved;
|
|
// search with hash is faster
|
|
resolved = custom_dlsym(&cache, symbol.hash, symbol.name);
|
|
if (resolved == 0) {
|
|
// but fuck apple they have relative path and rpath
|
|
resolved = custom_dlsym(&cache, symbol.lib, symbol.name);
|
|
}
|
|
*(uint64_t *)fix_at = (uint64_t)resolved;
|
|
|
|
printf("imports need to fix: (0x%x)%s at 0x%llx\n", symbol.hash,
|
|
symbol.name, fix_at);
|
|
printf(" from=%s\n", symbol.lib);
|
|
printf(" segment id=%d; offset=0x%llx;", symbol.segment_i, symbol.offset);
|
|
printf(" resolved=%llx(%p)\n", *(uint64_t*)fix_at, resolved);
|
|
}
|
|
|
|
// TODO: Reformat the region as per before, or leave as it
|
|
// for (int j = 0; j < npage_rw_fixed; j++) {
|
|
// uint64_t start_page = page_rw_fixed[j];
|
|
// vm_protect_func(mach_task_self_func(), start_page, 0x4000, 0, VM_PROT_READ);
|
|
// }
|
|
}
|
|
}
|
|
|
|
void test(struct libcache& cache) {
|
|
uint32_t libsystem_hash =
|
|
calculate_libname_hash(&cache, "/usr/lib/libSystem.B.dylib");
|
|
if (false) { // test search using name
|
|
void *printf_func =
|
|
custom_dlsym(&cache, "/usr/lib/libSystem.B.dylib", "_printf");
|
|
printf("Indirect search: Found=%p Expected=%p\n", printf_func, printf);
|
|
|
|
void *vm_protect_func =
|
|
custom_dlsym(&cache, "/usr/lib/libSystem.B.dylib", "_vm_protect");
|
|
printf("Indirect search: Found=%p Expected=%p\n", vm_protect_func,
|
|
vm_protect);
|
|
|
|
// using relative path
|
|
void *func_c_1 =
|
|
custom_dlsym(&cache, "./out/libb.dylib", "__Z15exported_from_cv");
|
|
printf("Indirect search: Found=%p Expected=%p\n", func_c_1,
|
|
exported_from_c);
|
|
|
|
// using rpath
|
|
void *func_c_2 =
|
|
custom_dlsym(&cache, "@rpath/libb.dylib", "__Z15exported_from_cv");
|
|
printf("Indirect search: Found=%p Expected=%p\n", func_c_2,
|
|
exported_from_c);
|
|
}
|
|
|
|
if (false) { // test search using hash of name
|
|
void *printf_func = custom_dlsym(&cache, libsystem_hash, "_printf");
|
|
printf("Indirect search: Found=%p Expected=%p\n", printf_func, printf);
|
|
|
|
void *vm_protect_func = custom_dlsym(&cache, libsystem_hash, "_vm_protect");
|
|
printf("Indirect search: Found=%p Expected=%p\n", vm_protect_func,
|
|
vm_protect);
|
|
|
|
void *realpath_func = custom_dlsym(&cache, libsystem_hash, "_realpath$DARWIN_EXTSN");
|
|
printf("Indirect search: Found=%p Expected=%p\n", realpath_func, realpath);
|
|
}
|
|
}
|