clean code and add comment

This commit is contained in:
nganhkhoa 2024-01-04 06:34:07 +07:00
parent 7a6a41b4d8
commit 263596b1a1
2 changed files with 97 additions and 96 deletions

View File

@ -472,7 +472,8 @@ func (mc *MachoContext) ReworkForObjc() {
main_offset := int(mc.entryoff)
var shellcode_offset int
if (mc.header.cputype & 0xff) == 12 {
isArm := (mc.header.cputype & 0xff) == 12
if isArm {
shellcode = []uint32{
0x10000008,
0, // x9 = (offset end of __DATA) - (offset shellcode)
@ -513,37 +514,11 @@ func (mc *MachoContext) ReworkForObjc() {
// fmt.Printf("// movz_data_end_offset=%x\n", movz_data_end_offset)
fmt.Printf("// lc_main_offset=%x\n", lc_main_offset)
} else {
// TODO: fix to work with offset larger than 0xffff
// shellcode = []uint32{
// 0x00058d4c,
// 0x66000000,
// 0, // offset
// 0x57c8014d,
// 0x41515256,
// 0x088b4d50,
// 0x41d1ff41,
// 0x5e5a5958,
// 0x488b4d5f,
// 0xe1ff4108,
// }
shellcode_x := []uint8{
shellcode_start := []uint8{
0x4c, 0x8d, 0x05, 0x00, 0x00, 0x00, 0x00,
0x49, 0xC7, 0xC1,
}
offset := []uint8{0x00, 0x00, 0x00, 0x00} // offset
shellcode_offset = text_start - 44
encode_movz := func(v int) {
for i := 0; i < 4; i++ {
offset[i] = uint8(v >> (i * 8))
}
}
// 7 is shellcode size to get RIP
encode_movz((data_end - text_start) + (44 - 7))
shellcode_end := []uint8{
0x4d, 0x01, 0xc8,
0x57,
@ -560,33 +535,59 @@ func (mc *MachoContext) ReworkForObjc() {
0x5e,
0x5f, 0x4d, 0x8b, 0x48, 0x08,
0x41, 0xff, 0xe1,
// pad to %4
0x00, 0x00,
}
shellcode_x = append(shellcode_x, offset...)
shellcode_x = append(shellcode_x, shellcode_end...)
offset := []uint8{0x00, 0x00, 0x00, 0x00} // offset
shellcode_size := len(shellcode_start) + len(offset) + len(shellcode_end)
for i := 0; i < len(shellcode_x)-2; i += 4 {
// could use buffer encoding, but for correctness,
// we do this by hand
encode_movz := func(v int) {
for i := 0; i < 4; i++ {
offset[i] = uint8(v >> (i * 8))
}
}
// ┌─────────────────┐
// │ │
// shellcode starts ────┼─────────────────┼───── │ │instruction
// │ │ │ │fetch RIP size
// RIP returns ────┼─────────────────┼───── ▲ │ │
// │ │ │ │
// │ │ │ │ shellcode length
// shellcode ends │ │ │ offset │
// __text ────┼─────────────────┼───── │ range │
// │ │ │ │ __DATA ends - __text
// │ │ │ │
// __DATA ends ────┼─────────────────┼───── ▼ │
// │ │
// │ │
// │ │
// │ │
// │ │
// └─────────────────┘
encode_movz((data_end - text_start) + (shellcode_size - len(shellcode_start)))
shellcode_offset = text_start - shellcode_size
shellcode_bytes := append(shellcode_start, offset...)
shellcode_bytes = append(shellcode_bytes, shellcode_end...)
for i := 0; i < len(shellcode_bytes); i += 4 {
val := 0
// little endian
val |= int(shellcode_x[i+0]) << 0
val |= int(shellcode_x[i+1]) << 8
val |= int(shellcode_x[i+2]) << 16
val |= int(shellcode_x[i+3]) << 24
val |= int(shellcode_bytes[i+0]) << 0
val |= int(shellcode_bytes[i+1]) << 8
val |= int(shellcode_bytes[i+2]) << 16
val |= int(shellcode_bytes[i+3]) << 24
shellcode = append(shellcode, uint32(val))
}
shellcode = append(shellcode, 0x0000e1ff)
// shellcode[10] = movz_data_end_offset
// shellcode[19] = movz_main_offset
fmt.Printf("// shellcode_offset=%x\n", shellcode_offset)
fmt.Printf("// main_offset=%x\n", main_offset)
fmt.Printf("// data_end=%x\n", data_end)
// fmt.Printf("// movz_calculate_offset=%x\n", movz_calculate_offset)
// fmt.Printf("// movz_shellcode_offset=%x\n", movz_shellcode_offset)
// fmt.Printf("// movz_main_offset=%x\n", movz_main_offset)
// fmt.Printf("// movz_data_end_offset=%x\n", movz_data_end_offset)
fmt.Printf("// lc_main_offset=%x\n", lc_main_offset)
}

View File

@ -389,29 +389,12 @@ void *get_export_trie(const void *header, uint32_t &size) {
const uint32_t offset = *((uint32_t *)ptr + 2);
size = *((uint32_t *)ptr + 3);
uint64_t offset_in_linkedit = (uint64_t)offset - linkedit_fileoffset;
printf("trie: %p\n",linkedit_vmaddr + slice + offset_in_linkedit );
if (linkedit_vmaddr + slice + offset_in_linkedit == 0x7ff888187f78) {
FILE* f = fopen("../scripts/libsystem_c.dylib", "wb");
// fwrite(header, size + linkedit_vmaddr + slice + offset_in_linkedit, 1, f);
// fwrite((char*)linkedit_vmaddr + slice + offset_in_linkedit, size, 1, f);
fclose(f);
}
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;
printf("trie: %p\n",linkedit_vmaddr + slice + offset_in_linkedit );
if (linkedit_vmaddr + slice + offset_in_linkedit == 0x7ff888187f78) {
FILE* f = fopen("../scripts/libsystem_c.bin", "wb");
fwrite((char*)linkedit_vmaddr + slice + offset_in_linkedit, size, 1, f);
fclose(f);
}
return (void *)(linkedit_vmaddr + slice + offset_in_linkedit);
}
if (cmd == LC_SEGMENT_64) {
@ -507,8 +490,22 @@ void *find_in_export_trie(const void *header, void *trie, char *& symbol) {
decode_uleb128(ptr, &func);
if (flag == 0x8 /*re-export*/) {
// this hits a re-export symbol but with another name
// usually, the re-export is the same name on another library
// but somehow, for system libraries, a lot of symbols are
// renamed and re-exported from another library
// probably this was to build wrappers and
// have custom platform optimizations
//
// example of these is _strlen in libsystem_c.dylib
// is re-exported from __platform_strlen in libplatform
//
// The purpose of using char*& is to change the symbol searching
// to another symbols and do it quickly using references (pointer)
//
// we return 0 so the dlsym continues to search,
// but with another symbol name because the symbol points to another string
symbol = ptr;
printf("re-export %s\n", symbol);
return 0;
}
return (void *)((char *)header + func);
@ -560,21 +557,34 @@ void *find_in_lib(struct libcache *cache, struct libcache_item *lib,
return direct;
}
// cannot find in directly exported trie, loop through all reexport libs
// printf("looking for symbol: %s\n", symbol);
return find_in_reexport(cache, lib, symbol);
}
// the current logic of dlsym is not correct, but it works for PoC
//
// dlsym searchs and match libraries based on the LC_DYLD_ID load command
// while for our PoC, we use the paths of libraries to search for them
//
// for performance reasons, we do not compare the paths as strings
// we instead use a simple hash to carry out comparision
// using hashes allows us to compare integers and would be faster
void *custom_dlsym(struct libcache *cache, uint32_t hash, const char *symbol) {
for (size_t i = 0; i < cache->size; i++) {
struct libcache_item *cache_lib = &cache->libs[i];
if (cache_lib->hash == hash) {
// if (custom_strcmp(symbol, "_strlen") == 0) {
// return find_in_lib(cache, cache_lib, "__platform_strlen");
// }
// read find_in_export_trie comments to know the use of char*&
//
// this code is for when the symbol searching references
// a previous item in search chain
//
// For example:
// searching for X in [A, B, C],
// C has X but it is a re-export from B with the name Y
// then we have to perform a search again from the top
// but with symbol Y
char** symbol_copy = (char**)&symbol;
void* func = find_in_lib(cache, cache_lib, *symbol_copy);
if (*symbol_copy != symbol) {
printf("symbol found %p %s %s\n", func, *symbol_copy, symbol);
func = find_in_lib(cache, cache_lib, *symbol_copy);
}
return func;
@ -901,12 +911,12 @@ void build_cache(struct libcache &cache, void *main) {
char* dyld_image_count_s = "__dyld_image_count";
int (*dyld_image_count_func)(void) = (dyld_image_count_t)find_in_export_trie(
libdyld, libdyld_export_trie, dyld_image_count_s);
char* dyld_get_image_header_s = "__dyld_get_image_header";
void *(*dyld_get_image_header_func)(int) =
(dyld_get_image_header_t)find_in_export_trie(libdyld, libdyld_export_trie,
dyld_get_image_header_s);
char* dyld_get_image_name_s = "__dyld_get_image_name";
char *(*dyld_get_image_name_func)(int) =
(dyld_get_image_name_t)find_in_export_trie(libdyld, libdyld_export_trie,
@ -1182,7 +1192,7 @@ void fix(struct libcache &cache) {
void volatile custom_initializer(int argc, const char *const argv[],
const char *const envp[],
const char *const apple[]) {
printf("run custom initializers %p\n", custom_initializer_i);
printf("[+] run custom initializers\n");
if (custom_initializer_i->cls != 0) {
// for Objective-C load
@ -1251,7 +1261,7 @@ void volatile custom_initializer(int argc, const char *const argv[],
free(custom_initializer_i->constructors);
}
printf("initializers completed\n");
printf("[+] initializers completed\n");
free(custom_initializer_i);
}
@ -1452,30 +1462,18 @@ void fix_initializer(struct libcache_item *libfixing, struct libcache &cache) {
// (note: __TEXT segment is aligned to the end of the page, free space in the
// middle)
//
// Below is the shellcode.
/*
adr x8, 0
# x9 = (offset end of __DATA) - (offset shellcode)
movz x9, #0x9999
add x8, x8, x9
stp x30, x8, [sp], #-0x10
stp x3, x2, [sp], #-0x10
stp x1, x0, [sp], #-0x10
# custom intializer
ldr x9, [x8]
blr x9
ldp x1, x0, [sp, #0x10]!
ldp x3, x2, [sp, #0x10]!
ldp x30, x8, [sp, #0x10]!
# original main
# link register is set so jump only
ldr x9, [x8, #8]
br x9
*/
// The shellcode is built using the ios-wrapper tool
// The idea is:
//
// push main arguments
// r8 = shellcode location
// r9 = offset from shellcode to __DATA end
// r8 = r8 + r9 -- get __DATA end address
// r9 = *r8 -- the first pointer is custom_initializer
// call r9
// r9 = *(r8 + 4) -- the second pointer is main function
// pop main arguments
// jump r9 -- do not call, return to dyld
void *header = libfixing->header;
const uint32_t magic = *(uint32_t *)header;
@ -1542,7 +1540,9 @@ void fix_initializer(struct libcache_item *libfixing, struct libcache &cache) {
uint64_t *dummy = (uint64_t *)(addr + slide + size);
dummy[0] = (uint64_t)custom_initializer;
dummy[1] = (uint64_t)(header) + bshield_data::main;
printf("add custom main-peg at %p\n", dummy);
printf("-- add custom main-peg at %p\n", dummy);
printf("-- custom initializer at %llx\n", dummy[0]);
printf("-- main function at %llx\n", dummy[1]);
} else if (custom_strcmp(name, "__LINKEDIT") == 0) {
linkedit_vmaddr = vmaddr;
linkedit_fileoffset = fileoffset;