2024年最新IDA Python 使用总结_idapython(1)，2024阿里巴巴面试参考指南
我们在看视频学习的时候，不能光动眼动脑不动手，比较科学的学习方法是在理解之后运用它们，这时候练手项目就很适合了，只是里面的项目比较多，水平也是参差不齐，大家可以挑自己能做的项目去练练。这是我花了几天的时间去把Python所有方向的技术点做的整理，形成各个领域的知识点汇总，它的用处就在于，你可以按照上面的知识点去找对应的学习资源，保证自己学得较为全面。④ Python基础入门、爬虫、web开发、大数
（1）Python所有方向的学习路线（新版）
这是我花了几天的时间去把Python所有方向的技术点做的整理，形成各个领域的知识点汇总，它的用处就在于，你可以按照上面的知识点去找对应的学习资源，保证自己学得较为全面。
最近我才对这些路线做了一下新的更新，知识体系更全面了。
（2）Python学习视频
包含了Python入门、爬虫、数据分析和web开发的学习视频，总共100多个，虽然没有那么全面，但是对于入门来说是没问题的，学完这些之后，你可以按照我上面的学习路线去网上找其他的知识资源进行进阶。
（3）100多个练手项目
我们在看视频学习的时候，不能光动眼动脑不动手，比较科学的学习方法是在理解之后运用它们，这时候练手项目就很适合了，只是里面的项目比较多，水平也是参差不齐，大家可以挑自己能做的项目去练练。
网上学习资料一大堆，但如果学到的知识不成体系，遇到问题时只是浅尝辄止，不再深入研究，那么很难做到真正的技术提升。
一个人可以走的很快，但一群人才能走的更远！不论你是正从事IT行业的老鸟或是对IT行业感兴趣的新人，都欢迎加入我们的的圈子（技术交流、学习资源、职场吐槽、大厂内推、面试辅导），让我们一起学习成长！
new_code = handle_one_branch(branch_address, new_code_ea)
ida_bytes.patch_bytes(new_code_ea, new_code)
# 当前 branch 去完混淆之后需要遍历代码找到 call 和 jmp 指令从而找到其他的 function 和 branch 。
ea = new_code_ea
while ea < new_code_ea + len(new_code):
idc.create_insn(ea)
if idc.print_insn_mnem(ea) == 'call':
call_target, need_handle = reloc.get(get_operand_value(ea, 0))
if need_handle: func_queue.put(call_target)
elif idc.print_insn_mnem(ea)[0] == 'j' and idc.get_operand_type(ea, 0) != idc.o_reg:
jcc_target, need_handle = reloc.get(get_operand_value(ea, 0))
if need_handle == True:
branch_queue.put(jcc_target)
ea += get_item_size(ea)
new_code_ea += len(new_code)
在完成代码去混淆之后需要对代码进行重定位，重定位的时候需要注意 jmp 指令长度的变化。
ea = new_code_start
while ea < new_code_ea:
idc.create_insn(ea)
mnem = idc.print_insn_mnem(ea)
if mnem == 'call':
call_target, need_handle = reloc.get(get_operand_value(ea, 0))
assert need_handle == False
ida_bytes.patch_bytes(ea, asm('call 0x%x' % (call_target), ea))
elif mnem[0] == 'j' and idc.get_operand_type(ea, 0) != idc.o_reg:
jcc_target, need_handle = reloc.get(get_operand_value(ea, 0))
assert need_handle == False
ida_bytes.patch_bytes(ea, asm('%s 0x%x' % (mnem, jcc_target), ea).ljust(idc.get_item_size(ea), b'\x90'))
elif mnem == 'pushf':
ida_bytes.patch_bytes(ea, b'\x90' \* 9)
ea += 9
continue
ea += get_item_size(ea)
最后去混淆后的 switch 不能被 ida 正常识别出来，具体原因是前面获取返回地址 eip 的函数被 patch 成了 `mov reg, xxx` 指令，导致其与编译器默认编译出的汇编不同（程序开启了 PIE，直接访问跳转表的地址 ida 不能正确识别），因此需要将这里的代码重新 patch 回去。
同时为了不影响原本程序中的数据，这里我将修复的跳转表放到了其他位置。另外还有两个字符串全局变量也移动到了正确位置。
new_jmp_table = (0xA6000 - 0x2D54, 0xA6000)
# 移动并修复跳转表
for eax in range(0x20):
jmp_target = (ida_bytes.get_dword(jmp_table[0] + eax \* 4) + jmp_table[1]) & 0xFFFFFFFF
new_jmp_target, need_handle = reloc.get(jmp_target)
assert need_handle == False
ida_bytes.patch_dword(new_jmp_table[0] + eax \* 4, (new_jmp_target - new_jmp_table[1]) & 0xFFFFFFFF)
need_patch_addr = 0x963D7
ida_bytes.patch_bytes(need_patch_addr, asm('call 0x900;add ebx, 0x%x' % (new_jmp_table[1] - (need_patch_addr + 5)), need_patch_addr)) # 修复指令
ida_bytes.patch_bytes(new_jmp_table[1] - 0x2d7a, ida_bytes.get_bytes(jmp_table[1] - 0x2d7a, 0x26)) # 复制字符串到正确位置
最终去混淆脚本如下：
from queue import *
import ida_bytes
from idc import *
import idc
from keystone import *
from capstone import *
asmer = Ks(KS_ARCH_X86, KS_MODE_32)
disasmer = Cs(CS_ARCH_X86, CS_MODE_32)
def disasm(machine_code, addr=0):
l = “”
for i in disasmer.disasm(machine_code, addr):
l += “{:8s} {};\n”.format(i.mnemonic, i.op_str)
return l.strip(‘\n’)
def asm(asm_code, addr=0):
l = b’’
for i in asmer.asm(asm_code, addr)[0]:
l += bytes([i])
return l
def print_asm(ea):
print(disasm(idc.get_bytes(ea, idc.get_item_size(ea)), ea))
class RelocDSU:
def \_\_init\_\_(self):
self.reloc = {}
def get(self, ea):
if ea not in self.reloc:
if idc.print_insn_mnem(ea) == 'jmp' and idc.get_operand_type(ea, 0) != idc.o_reg:
jmp_ea = idc.get_operand_value(ea, 0)
if idc.get_segm_name(jmp_ea) == '.got.plt':
self.reloc[ea] = ea
return self.reloc[ea], False
self.reloc[ea], need_handle = self.get(idc.get_operand_value(ea, 0))
return self.reloc[ea], need_handle
else:
self.reloc[ea] = ea
if self.reloc[ea] != ea: self.reloc[ea] = self.get(self.reloc[ea])[0]
return self.reloc[ea], idc.get_segm_name(self.reloc[ea]) == '.text'
def merge(self, ea, reloc_ea):
self.reloc[self.get(ea)[0]] = self.get(reloc_ea)[0]
reloc = RelocDSU()
class Block:
def __init__(self, start_ea, end_ea, imm, reg, call_target):
self.start_ea = start_ea
self.end_ea = end_ea
self.imm = imm
self.reg = reg
self.call_target = call_target
def mov_code(ea, new_code_ea):
return asm(disasm(idc.get_bytes(ea, idc.get_item_size(ea)), ea), new_code_ea)
def get_real_code(block, new_code_ea):
ea = block.call_target
while True:
if idc.print_insn_mnem(ea) == ‘cmp’:
reg = idc.print_operand(ea, 0)
imm = idc.get_operand_value(ea, 1)
if reg == block.reg and imm == block.imm:
ea += idc.get_item_size(ea)
break
ea += idc.get_item_size(ea)
# 在 cmp 判断找到对应位置后会依次执行 jnz，popa，popf 三条指令
assert idc.print_insn_mnem(ea) == 'jnz'
ea += idc.get_item_size(ea)
assert idc.print_insn_mnem(ea) == 'popa'
ea += idc.get_item_size(ea)
assert idc.print_insn_mnem(ea) == 'popf'
ea += idc.get_item_size(ea)
if idc.print_insn_mnem(ea) == 'pushf': # 第一种特殊情况，实际是 ret 指令。
return True, asm('ret')
new_code = b''
while True:
if idc.print_insn_mnem(ea) == 'jmp': # 第二种特殊情况，跳转过去可能还会有几条实际功能指令。
jmp_ea = idc.get_operand_value(ea, 0)
if idc.print_insn_mnem(jmp_ea) == 'pushf':
break
ea = jmp_ea
else:
code = mov_code(ea, new_code_ea)
new_code += code
new_code_ea += len(code)
ea += get_item_size(ea)
return False, new_code
def get_block(start_ea):
global imm, reg, call_target
mnem_list = [‘pushf’, ‘pusha’, ‘mov’, ‘call’, ‘pop’]
ea = start_ea
for i in range(5):
mnem = idc.print_insn_mnem(ea)
assert mnem == mnem_list[i]
if mnem == ‘mov’:
imm = idc.get_operand_value(ea, 1)
reg = idc.print_operand(ea, 0)
elif mnem == ‘call’:
call_target = idc.get_operand_value(ea, 0)
ea += idc.get_item_size(ea)
return Block(start_ea, ea, imm, reg, call_target)
def handle_one_branch(branch_address, new_code_ea):
new_code = b’’
ea = branch_address
while True:
try:
block = get_block(ea)
is_ret, real_code = get_real_code(block, new_code_ea)
reloc.merge(ea, new_code_ea)
ea = block.end_ea
new_code_ea += len(real_code)
new_code += real_code
if is_ret: break
except:
get_eip_func = {0x900: ‘ebx’, 0x435c: ‘eax’}
if idc.print_insn_mnem(ea) == ‘call’ and get_operand_value(ea, 0) in get_eip_func:
reloc.merge(ea, new_code_ea)
real_code = asm(‘mov %s, 0x%x’ % (get_eip_func[get_operand_value(ea, 0)], ea + 5), new_code_ea)
else:
if idc.print_insn_mnem(ea) == ‘jmp’ and idc.get_operand_type(ea, 0) != idc.o_reg:
reloc.merge(new_code_ea, ea)
else:
reloc.merge(ea, new_code_ea)
real_code = mov_code(ea, new_code_ea)
new_code += real_code
if real_code == asm('ret'): break
new_code_ea += len(real_code)
if idc.print_insn_mnem(ea) == 'jmp' and idc.get_operand_type(ea, 0) != idc.o_reg: # jmp reg is a swtich
jmp_ea = idc.get_operand_value(ea, 0)
if reloc.get(jmp_ea)[1] == False: break # 跳回之前的代码说明是个循环
ea = reloc.get(jmp_ea)[0]
else:
ea += get_item_size(ea)
return new_code
def solve():
entry_point = 0x48F4
new_code_start = 0x96150
new_code_ea = new_code_start
jmp_table = (0x892ac, 0x8c000) # [0x8c000 + (eax>>2) - 0x2d54] + 0x8c000
for _ in range(0x10000): idc.del_items(new_code_ea + _)
ida_bytes.patch_bytes(new_code_ea, 0x10000 \* b'\x90')
func_queue = Queue()
func_queue.put(entry_point)
while not func_queue.empty():
func_address = func_queue.get()
if reloc.get(func_address)[1] == False: continue
reloc.merge(func_address, new_code_ea)
branch_queue = Queue()
branch_queue.put(func_address)
if func_address == 0x4148: # 特判 0x4148 地址处的函数，读取跳转表。
assert new_code_ea == 0x963d0
for eax in range(0x20):
jmp_target = (ida_bytes.get_dword(jmp_table[0] + eax \* 4) + jmp_table[1]) & 0xFFFFFFFF
new_jmp_target, need_handle = reloc.get(jmp_target)
if need_handle: branch_queue.put(jmp_target)
while not branch_queue.empty():
branch_address = branch_queue.get()
new_code = handle_one_branch(branch_address, new_code_ea)
ida_bytes.patch_bytes(new_code_ea, new_code)
# 当前 branch 去完混淆之后需要遍历代码找到 call 和 jmp 指令从而找到其他的 function 和 branch 。
ea = new_code_ea
while ea < new_code_ea + len(new_code):
idc.create_insn(ea)
if idc.print_insn_mnem(ea) == 'call':
call_target, need_handle = reloc.get(get_operand_value(ea, 0))
if need_handle: func_queue.put(call_target)
elif idc.print_insn_mnem(ea)[0] == 'j' and idc.get_operand_type(ea, 0) != idc.o_reg:
jcc_target, need_handle = reloc.get(get_operand_value(ea, 0))
if need_handle == True:
branch_queue.put(jcc_target)
ea += get_item_size(ea)
new_code_ea += len(new_code)
ea = new_code_start
while ea < new_code_ea:
idc.create_insn(ea)
mnem = idc.print_insn_mnem(ea)
if mnem == 'call':
call_target, need_handle = reloc.get(get_operand_value(ea, 0))
assert need_handle == False
ida_bytes.patch_bytes(ea, asm('call 0x%x' % (call_target), ea))
elif mnem[0] == 'j' and idc.get_operand_type(ea, 0) != idc.o_reg:
jcc_target, need_handle = reloc.get(get_operand_value(ea, 0))
assert need_handle == False
ida_bytes.patch_bytes(ea, asm('%s 0x%x' % (mnem, jcc_target), ea).ljust(idc.get_item_size(ea), b'\x90'))
elif mnem == 'pushf':
ida_bytes.patch_bytes(ea, b'\x90' \* 9)
ea += 9
continue
ea += get_item_size(ea)
new_jmp_table = (0xA6000 - 0x2D54, 0xA6000)
# 移动并修复跳转表
for eax in range(0x20):
jmp_target = (ida_bytes.get_dword(jmp_table[0] + eax \* 4) + jmp_table[1]) & 0xFFFFFFFF
new_jmp_target, need_handle = reloc.get(jmp_target)
assert need_handle == False
ida_bytes.patch_dword(new_jmp_table[0] + eax \* 4, (new_jmp_target - new_jmp_table[1]) & 0xFFFFFFFF)
need_patch_addr = 0x963D7
ida_bytes.patch_bytes(need_patch_addr, asm('call 0x900;add ebx, 0x%x' % (new_jmp_table[1] - (need_patch_addr + 5)), need_patch_addr)) # 修复指令
ida_bytes.patch_bytes(new_jmp_table[1] - 0x2d7a, ida_bytes.get_bytes(jmp_table[1] - 0x2d7a, 0x26)) # 复制字符串到正确位置
for _ in range(0x10000): idc.del_items(new_code_ea + _)
idc.jumpto(new_code_start)
ida_funcs.add_func(new_code_start)
print("finish")
solve()
### 例题：SUSCTF2022 tttree
[附件下载链接]( )
首先将 `0x140010074` ，`0x140017EFA` ，`140018C67` 起始处的数据转换为汇编。
观察汇编，发现很多代码块之间相互跳转，因此先按照 `retn` 划分代码块。通过对代码块的观察，发现这些代码块按照 `call $+5;pop rax`（即 `E8 00 00 00 00 58` ） 的出现次数可以分为三种：
* 出现 0 次：
本质上是 `其它操作` + `retn` 。
* 出现 1 次：
这种代码块本质为 `其它操作` + `jmp target` ，注意 `其它操作` 中可能包含 branch 。
* 出现 2 次：
这个可以看做 2 个出现 1 次的代码块两个拼在一起，其中前面一个代码块去掉 `retn` 。执行完前面一个代码块后由于没有 `retn` ，因此 `target1` 留在栈中。执行第 2 个代码块跳转到 `target2` 执行 ，在 `target2` 代码块返回时会返回到 `target1` 。因此这种代码块本质上相当于 `其它操作` + `call target2` 且下一个要执行的代码块为 `target1` 。
我们定义代码块 `Block` 几个关键信息：
* `start_addr`：代码块的起始地址。
* `asm_list`：代码块的有效汇编，由于汇编指令可能包含 `[rip + xxx]` ，因此需要记录汇编指令的地址以便后续修正。
* `direct_next`：执行完此代码块后接下来要执行的代码块地址。
* `branch_list`：代码块中的所有条件跳转语句跳到的地址。
* `call_target`：代码块调用函数地址。
class Block:
def __init__(self, start_ea, asm_list, direct_next, branch_list, call_target):
self.start_ea = start_ea
self.asm_list = asm_list
self.direct_next = direct_next
self.branch_list = branch_list
self.call_target = call_target
def \_\_str\_\_(self):
return 'start\_ea: 0x%x\ndirect\_next: 0x%x\ncall\_target: 0x%x\nbranch\_list: %s\nasm\_list:\n%s\n' % (
0 if self.start_ea == None else self.start_ea,
0 if self.direct_next == None else self.direct_next,
0 if self.call_target == None else self.call_target,
str([hex(x) for x in self.branch_list]),
str('\n'.join([hex(addr) + ' ' + asm for addr, asm in self.asm_list]))
)
`get_block` 函数可以获取给定地址处的代码块并提取相关信息。代码块中可能有 `push xxx;pop xxx;` 这样的无意义指令，可以通过栈模拟来去除。
def get_block(start_ea):
ea = start_ea
stack = []
asm_list = []
branch_list = []
call_target = None
direct_next = None
while True:
idc.create_insn(ea)
mnem = idc.print_insn_mnem(ea)
# 处理混淆中跳转的情况
if mnem == 'pushfq':
ea += idc.get_item_size(ea)
assert idc.get_bytes(ea, idc.get_item_size(ea)) == b'\xE8\x00\x00\x00\x00'
ea += idc.get_item_size(ea)
jmp_base = ea
assert idc.print_insn_mnem(ea) == 'pop' and idc.get_operand_type(ea, 0) == o_reg
reg = idc.print_operand(ea, 0)
ea += idc.get_item_size(ea)
assert idc.print_insn_mnem(ea) == 'add' and idc.print_operand(ea, 0) == reg
assert idc.get_operand_type(ea, 1) == o_imm
jmp_target = (jmp_base + idc.get_operand_value(ea, 1)) & 0xFFFFFFFFFFFFFFFF
ea += idc.get_item_size(ea)
assert idc.get_bytes(ea, idc.get_item_size(ea)) == asm('mov [rsp + 0x10], %s' % reg, ea)
ea += idc.get_item_size(ea)
assert idc.print_insn_mnem(ea) == 'popfq'
ea += idc.get_item_size(ea)
assert idc.print_insn_mnem(ea) == 'pop' and idc.print_operand(ea, 0) == reg
assert len(stack) != 0 and stack[-1][0] == 'push' and stack[-1][1] == reg
stack.pop()
asm_list.pop()
assert len(stack) != 0 and stack[-1][0] == 'push' and stack[-1][1] == reg
stack.pop()
asm_list.pop()
ea += idc.get_item_size(ea)
if idc.print_insn_mnem(ea) == 'retn':
if direct_next == None:
direct_next = jmp_target
elif call_target == None:
call_target = jmp_target
asm_list.append((0, 'call 0x%x' % (call_target)))
else:
print("🤔🤔🤔🤔🤔🤔")
assert False
break
else:
assert call_target == None and direct_next == None
direct_next = jmp_target
continue
if mnem == 'push':
stack.append((mnem, idc.print_operand(ea, 0)))
elif mnem == 'pop':
if len(stack) != 0 and stack[-1][0] == 'push' and stack[-1][1] == idc.print_operand(ea, 0):
stack.pop()
asm_list.pop()
ea += idc.get_item_size(ea)
continue
else:
stack.clear()
else:
stack.clear()
asm_list.append((ea, disasm(idc.get_bytes(ea, idc.get_item_size(ea)), ea)))
if mnem == 'retn': break
if mnem[0] == 'j' and mnem != 'jmp' and idc.get_operand_type(ea, 0) != o_reg:
branch_list.append(idc.get_operand_value(ea, 0))
if mnem == 'jmp':
if idc.get_segm_name(idc.get_operand_value(ea, 0)) not in ['.text', '.aaa']:
break
else:
ea = idc.get_operand_value(ea, 0)
else:
ea += idc.get_item_size(ea)
return Block(start_ea, asm_list, direct_next, branch_list, call_target)
能够获取代码块信息之后就可以 bfs 函数以及函数中的所有分支，提取出汇编代码并写入 `newcode` 段。这里需要注意以下几点：
* 涉及 rip 的汇编指令不能只是简单把指令中的 `rip` 替换为对应的具体数值，因为有的指令立即数的长度被限制在 4 字节，直接替换成数值会溢出。一个比较好的解决方法是将 `rip` 替换为 `rip + (指令原本地址 - 指令当前地址)` 。这样借助 rip 寄存器扩大访问范围并且代码移动的距离不会超过 0x100000000 因此可以保证正确性。
* 如果 `block.direct_next` 对应的代码已经被去混淆了需要加上一条 jmp 指令跳转到已经去混淆的代码。
* 有的汇编指令 keystone 不支持汇编，比如 `bnd ret` ，需要特判。
while not func_queue.empty():
func_address = func_queue.get()
if reloc.get(func_address)[1] == False: continue
branch_queue = Queue()
branch_queue.put(func_address)
while not branch_queue.empty():
branch_address = branch_queue.get()
ea = branch_address
while True:
block = get_block(ea)
reloc.merge(ea, new_code_ea)
for addr, insn in block.asm_list:
insn = insn.replace('rip', 'rip - 0x%x' % (new_code_ea - addr))
if insn == 'bnd ret ;':
code = b'\xF2\xC3'
else:
code = asm(insn, new_code_ea)
ida_bytes.patch_bytes(new_code_ea, code)
if addr != 0: reloc.merge(addr, new_code_ea)
new_code_ea += len(code)
if block.call_target != None:
call_target, need_handle = reloc.get(block.call_target)
if need_handle: func_queue.put(call_target)
for branch_address in block.branch_list:
jcc_target, need_handle = reloc.get(branch_address)
if need_handle: branch_queue.put(jcc_target)
if block.direct_next == None: break
next_target, need_handle = reloc.get(block.direct_next)
if need_handle == False:
code = asm('jmp 0x%x' % (next_target), new_code_ea)
ida_bytes.patch_bytes(new_code_ea, code)
new_code_ea += len(code)
break
else:
ea = block.direct_next
最后对代码进行重定位，需要注意的是代码块中的有效指令中也可能有 call 指令，这里 call 调用的是一个类似 plt 表的结构，会直接跳转到导入表中的函数地址表指向的函数，需要特判这种情况。
ea = new_code_start
while ea < new_code_ea:
assert idc.create_insn(ea) != 0
mnem = idc.print_insn_mnem(ea)
if mnem == 'call':
call_target, need_handle = reloc.get(get_operand_value(ea, 0))
if need_handle == True:
if idc.print_insn_mnem(call_target) == 'jmp' and idc.get_segm_name(idc.get_operand_value(call_target, 0)) == '.idata':
ea += get_item_size(ea)
continue
else:
assert False
ida_bytes.patch_bytes(ea, asm('call 0x%x' % (call_target), ea).ljust(idc.get_item_size(ea), b'\x90'))
elif mnem[0] == 'j' and idc.get_operand_type(ea, 0) != idc.o_reg:
jcc_target, need_handle = reloc.get(get_operand_value(ea, 0))
assert need_handle == False
ida_bytes.patch_bytes(ea, asm('%s 0x%x' % (mnem, jcc_target), ea).ljust(idc.get_item_size(ea), b'\x90'))
ea += get_item_size(ea)
最后完整代码：
from queue import *
from idc import *
import idc
from keystone import *
from capstone import *
asmer = Ks(KS_ARCH_X86, KS_MODE_64)
disasmer = Cs(CS_ARCH_X86, CS_MODE_64)
def disasm(machine_code, addr=0):
l = “”
for i in disasmer.disasm(machine_code, addr):
l += “{:8s} {};\n”.format(i.mnemonic, i.op_str)
return l.strip(‘\n’)
def asm(asm_code, addr=0):
l = b’’
for i in asmer.asm(asm_code, addr)[0]:
l += bytes([i])
return l
class RelocDSU:
def \_\_init\_\_(self):
self.reloc = {}
def get(self, ea):
if ea not in self.reloc:
if idc.print_insn_mnem(ea) == 'jmp' and idc.get_operand_type(ea, 0) != idc.o_reg:
jmp_ea = idc.get_operand_value(ea, 0)
if idc.get_segm_name(jmp_ea) == '.idata':
self.reloc[ea] = ea
return self.reloc[ea], False
self.reloc[ea], need_handle = self.get(idc.get_operand_value(ea, 0))
return self.reloc[ea], need_handle
else:
self.reloc[ea] = ea
if self.reloc[ea] != ea: self.reloc[ea] = self.get(self.reloc[ea])[0]
return self.reloc[ea], idc.get_segm_name(self.reloc[ea]) in ['.text', '.aaa']
def merge(self, ea, reloc_ea):
# print((hex(ea), hex(reloc\_ea)))
self.reloc[self.get(ea)[0]] = self.get(reloc_ea)[0]
reloc = RelocDSU()
class Block:
def __init__(self, start_ea, asm_list, direct_next, branch_list, call_target):
self.start_ea = start_ea
self.asm_list = asm_list
self.direct_next = direct_next
self.branch_list = branch_list
self.call_target = call_target
def \_\_str\_\_(self):
return 'start\_ea: 0x%x\ndirect\_next: 0x%x\ncall\_target: 0x%x\nbranch\_list: %s\nasm\_list:\n%s\n' % (
0 if self.start_ea == None else self.start_ea,
0 if self.direct_next == None else self.direct_next,
0 if self.call_target == None else self.call_target,
str([hex(x) for x in self.branch_list]),
str('\n'.join([hex(addr) + ' ' + asm for addr, asm in self.asm_list]))
)
def get_block(start_ea):
ea = start_ea
stack = []
asm_list = []
branch_list = []
call_target = None
direct_next = None
while True:
idc.create_insn(ea)
mnem = idc.print_insn_mnem(ea)
# 处理混淆中跳转的情况
if mnem == 'pushfq':
ea += idc.get_item_size(ea)
assert idc.get_bytes(ea, idc.get_item_size(ea)) == b'\xE8\x00\x00\x00\x00'
ea += idc.get_item_size(ea)
jmp_base = ea
assert idc.print_insn_mnem(ea) == 'pop' and idc.get_operand_type(ea, 0) == o_reg
reg = idc.print_operand(ea, 0)
ea += idc.get_item_size(ea)
assert idc.print_insn_mnem(ea) == 'add' and idc.print_operand(ea, 0) == reg
assert idc.get_operand_type(ea, 1) == o_imm
jmp_target = (jmp_base + idc.get_operand_value(ea, 1)) & 0xFFFFFFFFFFFFFFFF
ea += idc.get_item_size(ea)
assert idc.get_bytes(ea, idc.get_item_size(ea)) == asm('mov [rsp + 0x10], %s' % reg, ea)
ea += idc.get_item_size(ea)
assert idc.print_insn_mnem(ea) == 'popfq'
ea += idc.get_item_size(ea)
assert idc.print_insn_mnem(ea) == 'pop' and idc.print_operand(ea, 0) == reg
assert len(stack) != 0 and stack[-1][0] == 'push' and stack[-1][1] == reg
stack.pop()
asm_list.pop()
assert len(stack) != 0 and stack[-1][0] == 'push' and stack[-1][1] == reg
stack.pop()
asm_list.pop()
ea += idc.get_item_size(ea)
if idc.print_insn_mnem(ea) == 'retn':
if direct_next == None:
direct_next = jmp_target
elif call_target == None:
call_target = jmp_target
asm_list.append((0, 'call 0x%x' % (call_target)))
else:
print("🤔🤔🤔🤔🤔🤔")
assert False
break
else:
assert call_target == None and direct_next == None
direct_next = jmp_target
continue
if mnem == 'push':
stack.append((mnem, idc.print_operand(ea, 0)))
elif mnem == 'pop':
if len(stack) != 0 and stack[-1][0] == 'push' and stack[-1][1] == idc.print_operand(ea, 0):
stack.pop()
asm_list.pop()
ea += idc.get_item_size(ea)
continue
else:
stack.clear()
else:
stack.clear()
asm_list.append((ea, disasm(idc.get_bytes(ea, idc.get_item_size(ea)), ea)))
if mnem == 'retn': break
if mnem[0] == 'j' and mnem != 'jmp' and idc.get_operand_type(ea, 0) != o_reg:
branch_list.append(idc.get_operand_value(ea, 0))
if mnem == 'jmp':
if idc.get_segm_name(idc.get_operand_value(ea, 0)) not in ['.text', '.aaa']:
break
else:
ea = idc.get_operand_value(ea, 0)
else:
ea += idc.get_item_size(ea)
return Block(start_ea, asm_list, direct_next, branch_list, call_target)
entry_point = 0x1400133B7
new_code_start = 0x14001D000
def solve():
for i in range(0x10000):
idc.set_name(new_code_start + i, ‘’)
idc.del_items(new_code_start + i)
ida_bytes.patch_bytes(new_code_start, b’\x90’ * 0x10000)
func_queue = Queue()
func_queue.put(entry_point)
new_code_ea = new_code_start
while not func_queue.empty():
func_address = func_queue.get()
if reloc.get(func_address)[1] == False: continue
branch_queue = Queue()
branch_queue.put(func_address)
while not branch_queue.empty():
branch_address = branch_queue.get()
ea = branch_address
while True:
block = get_block(ea)
reloc.merge(ea, new_code_ea)
for addr, insn in block.asm_list:
insn = insn.replace('rip', 'rip - 0x%x' % (new_code_ea - addr))
if insn == 'bnd ret ;':
code = b'\xF2\xC3'
else:
code = asm(insn, new_code_ea)
ida_bytes.patch_bytes(new_code_ea, code)
if addr != 0: reloc.merge(addr, new_code_ea)
new_code_ea += len(code)
if block.call_target != None:
call_target, need_handle = reloc.get(block.call_target)
if need_handle: func_queue.put(call_target)
for branch_address in block.branch_list:
jcc_target, need_handle = reloc.get(branch_address)
if need_handle: branch_queue.put(jcc_target)
if block.direct_next == None: break
next_target, need_handle = reloc.get(block.direct_next)
if need_handle == False:
code = asm('jmp 0x%x' % (next_target), new_code_ea)
ida_bytes.patch_bytes(new_code_ea, code)
new_code_ea += len(code)
break
else:
ea = block.direct_next
ea = new_code_start
while ea < new_code_ea:
assert idc.create_insn(ea) != 0
mnem = idc.print_insn_mnem(ea)
if mnem == 'call':
call_target, need_handle = reloc.get(get_operand_value(ea, 0))
if need_handle == True:
if idc.print_insn_mnem(call_target) == 'jmp' and idc.get_segm_name(idc.get_operand_value(call_target, 0)) == '.idata':
ea += get_item_size(ea)
continue
else:
assert False
ida_bytes.patch_bytes(ea, asm('call 0x%x' % (call_target), ea).ljust(idc.get_item_size(ea), b'\x90'))
elif mnem[0] == 'j' and idc.get_operand_type(ea, 0) != idc.o_reg:
jcc_target, need_handle = reloc.get(get_operand_value(ea, 0))
assert need_handle == False
ida_bytes.patch_bytes(ea, asm('%s 0x%x' % (mnem, jcc_target), ea).ljust(idc.get_item_size(ea), b'\x90'))
ea += get_item_size(ea)
for i in range(0x10000): idc.del_items(new_code_start + i)
idc.jumpto(new_code_start)
idc.add_func(new_code_start)
print("finish")
solve()
## 去花指令
花指令目前没有太好的去除办法，但是同一题目中花指令种类和变化都是有限的，也就是说我们可以将题目中所有花指令的类型总结出来，然后分别编写相应的查找和处理规则。
### 例题：看雪CTF2019 圆圈舞DancingCircle
[附件下载链接]( )
用IDA打开DancingCircle，按G输入 `0x401f58` 跳转至核心函数，发现有大量花指令。因此需要借助 ida python 脚本正则表达式匹配去除。
分析汇编代码，发现花指令有如下几类：
#### call 花指令
* call + pop
例如 0x00401F9B 处的花指令
另外还有 push eax + call + pop eax + pop eax 类型的。
* call + add esp, 4
例如 0x00401F62 处的花指令
* call + add [esp], 6 + retn
例如 0x00401FA3 处的花指令
#### jx + jnx 花指令
例如 0x00402D67 处的花指令
这类花指令可以做如下检测：
* 两个跳转指令的第一个字节相差 1 且较小的那个是偶数。
* 前一个跳转的立即数比后一个多 2 。
#### fake jmp 花指令
例如 0x00401FB2 这处花指令：
这里有很多跳转，但分析后发现这些跳转都可以忽略。由于这一类花指令比较单一，因此直接匹配特征即可。
#### stx + jx 花指令
例如 0x0040261F 和 0x004026D7 两处花指令：
此类花指令本质是通过设置标志寄存器的值使得满足后面的条件跳转。由于此类指令较少，直接匹配特征即可。
import regex as re
from idc import *
import idc
from keystone import *
from capstone import *
asmer = Ks(KS_ARCH_X86, KS_MODE_32)
disasmer = Cs(CS_ARCH_X86, CS_MODE_32)
def disasm(machine_code, addr=0):
l = “”
for i in disasmer.disasm(machine_code, addr):
l += “{:8s} {};\n”.format(i.mnemonic, i.op_str)
return l.strip(‘\n’)
def asm(asm_code, addr=0):
l = b’’
for i in asmer.asm(asm_code, addr)[0]:
l += bytes([i])
return l
def check_call_to_jmp(call_insn_addr):
call_target = idc.get_operand_value(call_insn_addr, 0)
if call_target not in range(start_ea, end_ea): return None
idc.create_insn(call_target)
if ida_bytes.get_bytes(call_target, idc.get_item_size(call_target)) == asm(‘add esp, 4’, call_target):
return call_target + idc.get_item_size(call_target)
if idc.print_insn_mnem(call_target) == ‘pop’:
return call_target + idc.get_item_size(call_target)
insn = disasm(ida_bytes.get_bytes(call_target, idc.get_item_size(call_target)), call_target)
if ‘[esp],’ in insn and (‘add’ in insn or ‘sub’ in insn) and idc.get_operand_type(call_target, 1) == o_imm:
idc.create_insn(call_target + idc.get_item_size(call_target))
if idc.print_insn_mnem(call_target + idc.get_item_size(call_target)) == ‘retn’:
return (call_insn_addr + 5 + (1 if idc.print_insn_mnem(call_target) == ‘add’ else -1) * idc.get_operand_value(call_target, 1)) & 0xFFFFFFFF
return None
def check_jcc_to_jmp(jcc_insn_addr):
code1 = ida_bytes.get_bytes(jcc_insn_addr, idc.get_item_size(jcc_insn_addr))
next_insn_addr = jcc_insn_addr + idc.get_item_size(jcc_insn_addr)
idc.create_insn(next_insn_addr)
code2 = ida_bytes.get_bytes(next_insn_addr, idc.get_item_size(next_insn_addr))
if abs(code1[0] - code2[0]) == 1 and min(code1[0], code2[0]) % 2 == 0 and idc.get_operand_value(jcc_insn_addr, 0) == idc.get_operand_value(next_insn_addr, 0):
return idc.get_operand_value(jcc_insn_addr, 0)
code = ida_bytes.get_bytes(jcc_insn_addr, 12)
print("bbbbb")
pattern_list = [
re.compile(rb"(?s)\x7C\x03\xEB\x03.\x74\xFB"),
re.compile(rb"(?s)\xEB\x07.\xEB\x01.\xEB\x04.\xEB\xF8."),
re.compile(rb"(?s)\xEB\x01.")
]
for pattern in pattern_list:
match = re.match(pattern, code)
if match != None and match.span()[1] != 0:
return jcc_insn_addr + match.span()[1]
return None
感谢每一个认真阅读我文章的人，看着粉丝一路的上涨和关注，礼尚往来总是要有的：
① 2000多本Python电子书（主流和经典的书籍应该都有了）
② Python标准库资料（最全中文版）
③ 项目源码（四五十个有趣且经典的练手项目及源码）
④ Python基础入门、爬虫、web开发、大数据分析方面的视频（适合小白学习）
⑤ Python学习路线图（告别不入流的学习）
网上学习资料一大堆，但如果学到的知识不成体系，遇到问题时只是浅尝辄止，不再深入研究，那么很难做到真正的技术提升。
一个人可以走的很快，但一群人才能走的更远！不论你是正从事IT行业的老鸟或是对IT行业感兴趣的新人，都欢迎加入我们的的圈子（技术交流、学习资源、职场吐槽、大厂内推、面试辅导），让我们一起学习成长！
开放原子开发者工作坊旨在鼓励更多人参与开源活动，与志同道合的开发者们相互交流开发经验、分享开发心得、获取前沿技术趋势。工作坊有多种形式的开发者活动，如meetup、训练营等，主打技术交流，干货满满，真诚地邀请各位开发者共同参与！
更多推荐
5
0- 0
已为社区贡献7条内容
所有评论(0)