1
#!/usr/bin/env python3
2
"""OurNotes 同构 APK 一把梭：python decrypt_apk.py input.apk [-o output]。 by MetaMiku"""
3
import argparse
4
import hashlib
5
import io
6
import struct
7
import zipfile
8
import zlib
9
from pathlib import Path
10

11
from capstone import Cs, CS_ARCH_ARM64, CS_MODE_LITTLE_ENDIAN
12
from Crypto.Cipher import AES
13
from elftools.elf.elffile import ELFFile
14

15
IV_BYTES = bytes(range(2, 18))
16
META_MAGIC = 0x12724394
17
TABLE_STRIDE = 80
18
SELECT = {'libanort': (0, 2), 'libanogs': (0, 2, 3, 4)}
19

20

21
def u32(data, offset=0):
22
    return struct.unpack_from('<I', data, offset)[0]
23

24

25
def find_all(data, needle):
26
    out, start = [], 0
27
    while True:
28
        pos = data.find(needle, start)
29
        if pos < 0:
30
            return out
31
        out.append(pos)
32
        start = pos + 1
33

34

35
class Elf:
36
    def __init__(self, data):
37
        self.data = data
38
        self.elf = ELFFile(io.BytesIO(data))
39
        assert self.elf.elfclass == 64 and self.elf.little_endian
40
        assert self.elf['e_machine'] == 'EM_AARCH64'
41
        self.loads = [p for p in self.elf.iter_segments() if p['p_type'] == 'PT_LOAD']
42
        text = self.elf.get_section_by_name('.text')
43
        self.text_base = text['sh_addr']
44
        off = self.try_offset(text['sh_addr'], text['sh_size'])
45
        self.words = struct.unpack(f"<{text['sh_size'] // 4}I",
46
                                   data[off:off + text['sh_size']])
47

48
    def try_offset(self, va, size=1):
49
        hits = [p['p_offset'] + va - p['p_vaddr'] for p in self.loads
50
                if p['p_vaddr'] <= va and va + size <= p['p_vaddr'] + p['p_filesz']]
51
        return hits[0] if len(hits) == 1 else None
52

53
    def offset(self, va, size=1):
54
        off = self.try_offset(va, size)
55
        assert off is not None, f'ELF 地址越界：0x{va:x}+0x{size:x}'
56
        return off
57

58
    def read(self, va, size):
59
        off = self.offset(va, size)
60
        return self.data[off:off + size]
61

62
    def va_of(self, offset):
63
        for section in self.elf.iter_sections():
64
            if section['sh_type'] == 'SHT_NOBITS' or not section['sh_addr']:
65
                continue
66
            if section['sh_offset'] <= offset < section['sh_offset'] + section['sh_size']:
67
                return section['sh_addr'] + offset - section['sh_offset']
68
        return None
69

70

71
def movz_movk(elf):
72
    """把 32 位 MOVZ LSL#0 + MOVK LSL#16 还原成 (高半地址, 常量)。"""
73
    pairs = []
74
    for i, word in enumerate(elf.words):
75
        if (word & 0x7F800000) != 0x72800000 or ((word >> 21) & 3) != 1:
76
            continue
77
        reg, high = word & 31, (word >> 5) & 0xFFFF
78
        for j in range(i - 1, max(-1, i - 8), -1):
79
            prev = elf.words[j]
80
            if (prev & 0x7F800000) == 0x52800000 and ((prev >> 21) & 3) == 0 \
81
                    and (prev & 31) == reg:
82
                pairs.append((elf.text_base + i * 4,
83
                              ((prev >> 5) & 0xFFFF) | (high << 16)))
84
                break
85
    return pairs
86

87

88
def crypt(data, seed, iv, aes_start, xor_start, encrypt=False):
89
    result = bytearray(data)
90
    table = bytes(i ^ (((seed >> 16) & 255) or 0x87) for i in range(256))
91
    if aes_start is None:
92
        return result.translate(table)
93
    assert 0 <= aes_start and aes_start + 0x4000 <= xor_start <= len(data)
94
    for pos in range(aes_start, aes_start + 0x4000, 0x800):
95
        aes = AES.new(f'{seed:08x}{seed:08x}'.encode(), AES.MODE_CBC, iv)
96
        transform = aes.encrypt if encrypt else aes.decrypt
97
        result[pos:pos + 0x800] = transform(data[pos:pos + 0x800])
98
    for pos in range(xor_start, len(data), 0x10000):
99
        end = pos + 0x4000 if len(data) - pos >= 0x10000 else len(data)
100
        result[pos:end] = result[pos:end].translate(table)
101
    return result
102

103

104
def check_code(data, address):
105
    instructions = list(Cs(CS_ARCH_ARM64, CS_MODE_LITTLE_ENDIAN).disasm_lite(data, address))
106
    assert sum(i[1] for i in instructions) == len(data), f'AES 代码校验失败：0x{address:x}'
107
    assert any(i[2] == 'ret' for i in instructions), '未找到函数返回'
108

109

110
def check_metadata(data):
111
    assert data[:8] == b'\0\0\0\0\x18\0\0\0'
112
    regions = [struct.unpack_from('<III', data, 8 + i * 12) for i in range(31)]
113
    end = 0x17C
114
    for offset, size, count in regions:
115
        assert end <= offset <= offset + size <= len(data), 'metadata 分区越界'
116
        end = offset + size
117
    assert regions[0][0] == 0x17C and end == len(data)
118
    offset, size, count = regions[0]
119
    assert count > 1 and size == count * 4
120
    indices = struct.unpack_from(f'<{count}I', data, offset)
121
    base, size, _ = regions[1]
122
    assert indices[0] == 0 and indices[-1] == size
123
    for start, end in zip(indices, indices[1:]):
124
        assert 0 <= start <= end <= size
125
        data[base + start:base + end].decode('utf-8')
126
    base, size, _ = regions[2]
127
    names = data[base:base + size]
128
    assert b'mscorlib.dll\0' in names
129
    for region, stride, fields in [(3, 24, (0,)), (4, 20, (0,)), (5, 32, (0,)),
130
                                   (10, 12, (0,)), (11, 12, (0,)),
131
                                   (19, 82, (0, 4)), (20, 36, (0,))]:
132
        offset, size, count = regions[region]
133
        assert size == stride * count, f'metadata 表 {region} 布局改变'
134
        for pos in range(offset, offset + size, stride):
135
            for field in fields:
136
                index = u32(data, pos + field)
137
                end = names.find(b'\0', index)
138
                assert 0 <= index <= end < len(names)
139
                names[index:end].decode('utf-8')
140
    offset, size, count = regions[19]
141
    for field, region in [(0x3A, 5), (0x3C, 4), (0x3E, 11), (0x40, 3),
142
                          (0x42, 15), (0x44, 17), (0x46, 16), (0x48, 18)]:
143
        total = sum(struct.unpack_from('<H', data, pos + field)[0]
144
                    for pos in range(offset, offset + size, 82))
145
        assert total == regions[region][2], f'metadata 表 {region} 成员总数不符'
146
    print(f'metadata：{regions[19][2]} 类型，{regions[5][2]} 方法，{regions[20][2]} 程序集')
147

148

149
def row_shape(table_data, target, row):
150
    """80 字节行结构：va/off/size 在 +0x10，MD5 十六进制在 +0x20；先只看形状。"""
151
    if row < 0 or row + 64 > len(table_data):
152
        return None
153
    va, off, size, _ = struct.unpack_from('<4I', table_data, row + 16)
154
    if not (0x1000 < va < 0x08000000 and 0 < size < 0x08000000):
155
        return None
156
    if target.try_offset(va, size) != off:
157
        return None
158
    md5 = table_data[row + 32:row + 64]
159
    if not all(c in b'0123456789abcdef' for c in md5):
160
        return None
161
    return va, off, size, md5
162

163

164
def table_row(table_data, target, row):
165
    """形状对了还不够，MD5 必须和实际段字节一致。"""
166
    shape = row_shape(table_data, target, row)
167
    if shape is None:
168
        return None
169
    va, off, size, md5 = shape
170
    if hashlib.md5(target.data[off:off + size]).hexdigest().encode() != md5:
171
        return None
172
    return va, off, size
173

174

175
def main():
176
    parser = argparse.ArgumentParser(description=__doc__)
177
    parser.add_argument('apk', type=Path)
178
    parser.add_argument('-o', '--output', type=Path)
179
    args = parser.parse_args()
180
    if not __debug__:
181
        raise SystemExit('请勿用 python -O 禁用校验')
182
    output = args.output or args.apk.with_name(args.apk.stem + '-decrypted')
183
    paths = [f'lib/arm64-v8a/{name}.so' for name in ('libanort', 'libanogs', 'libil2cpp')]
184
    paths += ['assets/bin/Data/Managed/Metadata/global-metadata.dat']
185
    with zipfile.ZipFile(args.apk) as apk:
186
        assert all(apk.namelist().count(p) == 1 for p in paths), 'APK 文件缺失或重复'
187
        anort, anogs, il2cpp = [Elf(apk.read(p)) for p in paths[:3]]
188
        metadata = apk.read(paths[3])
189

190
    # 1. 配置表：名字地址往前 16 字节
191
    needle = b'libil2cpp.so'.ljust(32, b'\0')
192
    config = None
193
    for pos in find_all(anort.data, needle):
194
        va = anort.va_of(pos)
195
        if va is None or va < 16 or anort.try_offset(va - 16, 0x224) is None:
196
            continue
197
        candidate = anort.read(va - 16, 0x224)
198
        init, saved = struct.unpack_from('<II', candidate, 80)
199
        if candidate[16:48] == needle and saved == 32 and il2cpp.try_offset(init, saved) is not None:
200
            config = candidate
201
            break
202
    assert config, '未找到 libil2cpp 配置表'
203
    print(f'配置表：init=0x{struct.unpack_from("<I", config, 80)[0]:x}')
204

205
    # 区域参数，后面解密和基值校验都要用
206
    plan = []
207
    previous_end = 0
208
    for i in range(8):
209
        entry = config[132 + i * 52:184 + i * 52]
210
        if not any(entry):
211
            continue
212
        name = entry[:16].split(b'\0')[0].decode('ascii')
213
        va, off, size = struct.unpack_from('<III', entry, 16)
214
        assert u32(entry, 48) == 1, f'{name} 解密模式改变'
215
        assert size and il2cpp.offset(va, size) == off and previous_end <= va
216
        previous_end = va + size
217
        plan.append((name, va, off, size, 0x2000 - (va & 0xFFF) if size >= 0x100000 else None))
218
    text = [item for item in plan if item[0] == '.text'][0]
219
    _, text_va, text_off, text_size, text_skip = text
220
    assert text_skip is not None and text_size >= text_skip + 0x4000
221
    cipher = il2cpp.data[text_off + text_skip:text_off + text_skip + 0x4000]
222

223
    # 2. metadata seed：锚定头部标记，同函数窗口内取 seed，用分区表验
224
    pairs = movz_movk(anort)
225
    magic = [address for address, value in pairs if value == META_MAGIC]
226
    assert magic, '未找到 metadata 头部标记 0x12724394'
227
    near = [value for address, value in pairs
228
            if magic[0] - 0x60 <= address <= magic[0] + 0x180 and value != META_MAGIC]
229
    metadata_seed = None
230
    for seed in dict.fromkeys(near):
231
        xor = ((seed >> 16) & 255) or 0x87
232
        head = bytes(byte ^ xor for byte in metadata[8:0x17C])
233
        regions = [struct.unpack_from('<III', head, i * 12) for i in range(31)]
234
        end = 0x17C
235
        if regions[0][0] != 0x17C:
236
            continue
237
        for offset, size, count in regions:
238
            if not (end <= offset <= offset + size <= len(metadata)):
239
                break
240
            end = offset + size
241
        else:
242
            if end == len(metadata):
243
                metadata_seed = seed
244
                break
245
    assert metadata_seed is not None, '未找到可用的 metadata seed'
246
    print(f'metadata seed：0x{metadata_seed:08x}')
247

248
    # 3. IV：唯一常量
249
    assert anort.data.count(IV_BYTES) == 1, 'IV 常量不唯一或缺失'
250
    iv = IV_BYTES
251
    print(f'IV：{iv.hex()}')
252

253
    # 4. 摘要表：按行结构扫描，实际字节验算后分两张表
254
    tables = {}
255
    for section in anort.elf.iter_sections():
256
        if section['sh_type'] == 'SHT_NOBITS' or not section['sh_flags'] & 0x2 \
257
                or section['sh_flags'] & 0x4:
258
            continue
259
        start, stop = section['sh_offset'], section['sh_offset'] + section['sh_size']
260
        for base in range(start, stop - TABLE_STRIDE * 5, 4):
261
            if base >= TABLE_STRIDE and any(
262
                    row_shape(anort.data, elf, base - TABLE_STRIDE) for elf in (anort, anogs)):
263
                continue
264
            for name, elf in (('libanort', anort), ('libanogs', anogs)):
265
                if all(table_row(anort.data, elf, base + i * TABLE_STRIDE)
266
                       for i in SELECT[name]):
267
                    tables.setdefault(name, set()).add(anort.va_of(base))
268
    for name in SELECT:
269
        assert len(tables.get(name, ())) == 1, f'{name} 摘要表定位失败：{tables.get(name)}'
270
    tables = {name: tables[name].copy().pop() for name in SELECT}
271
    print('摘要表：' + '，'.join(f'{name}=0x{value:x}' for name, value in tables.items()))
272

273
    # 5. 分量：按算法常量取行，CRC32(MD5) 异或
274
    component = 0
275
    for name, elf in (('libanort', anort), ('libanogs', anogs)):
276
        base = anort.offset(tables[name], TABLE_STRIDE * 5)
277
        for index in SELECT[name]:
278
            row = base + index * TABLE_STRIDE
279
            va, off, size = struct.unpack_from('<III', anort.data, row + 16)
280
            component ^= zlib.crc32(hashlib.md5(elf.data[off:off + size]).hexdigest().encode())
281
    print(f'摘要分量：0x{component:08x}')
282

283
    # 6. 状态基值：写栈槽指令 + 函数窗口内唯一的 MOVZ/MOVK
284
    candidates = {}
285
    for i, word in enumerate(anort.words):
286
        if word & 0xFFF00000 != 0xB9000000 or (word >> 10) & 0xFFF != 1 \
287
                or (word >> 5) & 31 != 31:
288
            continue
289
        reg = word & 31
290
        low = high = None
291
        low_n = high_n = 0
292
        for j in range(max(0, i - 0x600 // 4), i + 1):
293
            prev = anort.words[j]
294
            if (prev & 0x7F800000) == 0x52800000 and ((prev >> 21) & 3) == 0 \
295
                    and (prev & 31) == reg:
296
                low_n, low = low_n + 1, prev
297
            elif (prev & 0x7F800000) == 0x72800000 and ((prev >> 21) & 3) == 1 \
298
                    and (prev & 31) == reg:
299
                high_n, high = high_n + 1, prev
300
        if low_n == 1 and high_n == 1:
301
            candidates[anort.text_base + i * 4] = \
302
                ((low >> 5) & 0xFFFF) | (((high >> 5) & 0xFFFF) << 16)
303

304
    def text_ok(seed):
305
        plain = b''.join(
306
            AES.new(f'{seed:08x}{seed:08x}'.encode(), AES.MODE_CBC, iv).decrypt(
307
                cipher[pos:pos + 0x800])
308
            for pos in range(0, 0x4000, 0x800))
309
        try:
310
            check_code(plain, text_va + text_skip)
311
            return True
312
        except AssertionError:
313
            return False
314

315
    seeds = [(site, value ^ component) for site, value in candidates.items()
316
             if text_ok(value ^ component)]
317
    assert len(seeds) == 1, f'状态基值候选定位失败：{[hex(s) for s, _ in seeds]}'
318
    site, seed = seeds[0]
319
    print(f'状态基值：0x{candidates[site]:08x}（写栈槽 0x{site:x}）；code seed：0x{seed:08x}')
320

321
    # 7. 解密 libil2cpp：恢复初始化入口，再逐区域 AES + XOR
322
    init, saved_size = struct.unpack_from('<II', config, 80)
323
    offset = il2cpp.offset(init, saved_size)
324
    result = bytearray(il2cpp.data)
325
    result[offset:offset + saved_size] = config[48:80]
326
    for name, va, off, size, skip in plan:
327
        source = result[off:off + size]
328
        plain = crypt(source, seed, iv, skip, skip + 0x10000 if skip is not None else 0)
329
        assert crypt(plain, seed, iv, skip, skip + 0x10000 if skip is not None else 0,
330
                     True) == source
331
        if name in ('.text', 'il2cpp'):
332
            check_code(plain[skip:skip + 0x4000], va + skip)
333
        result[off:off + size] = plain
334
        print(f'解密区域：{name} RVA=0x{va:x} size=0x{size:x}')
335
    assert b'global-metadata.dat\0' in result
336
    entry = il2cpp.offset(init, 0x100)
337
    check_code(result[entry:entry + 0x100], init)
338

339
    # 8. 解密 metadata：XOR + AES + 周期 XOR，再清头部
340
    decoded = crypt(metadata, metadata_seed, iv, 0x1000, 0x11000)
341
    decoded[8:0x1000] = crypt(decoded[8:0x1000], metadata_seed, iv, None, 0)
342
    restored = crypt(decoded, metadata_seed, iv, 0x1000, 0x11000, True)
343
    restored[8:0x1000] = crypt(restored[8:0x1000], metadata_seed, iv, None, 0)
344
    assert restored == metadata
345
    decoded[:4], decoded[5:8] = bytes(4), bytes(3)
346
    check_metadata(decoded)
347

348
    output.mkdir(parents=True, exist_ok=True)
349
    for name, data in [('libil2cpp.so', result), ('global-metadata.dat', decoded)]:
350
        (output / name).write_bytes(data)
351
        print(f'{output / name}\n  MD5={hashlib.md5(data).hexdigest()}')
352

353

354
if __name__ == '__main__':
355
    main()