2021-12-25 22:38:24 +01:00
|
|
|
from random import randint
|
|
|
|
|
from time import time
|
|
|
|
|
from math import ceil
|
|
|
|
|
from Cryptodome.Hash import MD5
|
|
|
|
|
from Cryptodome.Cipher import Blowfish, AES
|
2021-12-30 23:54:30 +01:00
|
|
|
from requests.models import HTTPError
|
2021-12-31 15:53:57 +01:00
|
|
|
from tqdm import tqdm
|
2021-12-25 22:38:24 +01:00
|
|
|
from utils.utils import create_requests_session
|
|
|
|
|
|
2021-12-27 16:48:52 +01:00
|
|
|
class APIError(Exception):
|
|
|
|
|
def __init__(self, type, msg, payload):
|
|
|
|
|
self.type = type
|
|
|
|
|
self.msg = msg
|
|
|
|
|
self.payload = payload
|
|
|
|
|
def __str__(self):
|
|
|
|
|
return ', '.join((self.type, self.msg, str(self.payload)))
|
|
|
|
|
|
2021-12-25 22:38:24 +01:00
|
|
|
class DeezerAPI:
|
|
|
|
|
def __init__(self, exception, client_id, client_secret, bf_secret, track_url_key):
|
|
|
|
|
self.gw_light_url = 'https://www.deezer.com/ajax/gw-light.php'
|
|
|
|
|
self.api_token = ''
|
|
|
|
|
self.exception = exception
|
|
|
|
|
self.client_id = client_id
|
|
|
|
|
self.client_secret = client_secret
|
2021-12-27 16:48:52 +01:00
|
|
|
|
2021-12-25 22:38:24 +01:00
|
|
|
self.legacy_url_cipher = AES.new(track_url_key.encode('ascii'), AES.MODE_ECB)
|
|
|
|
|
self.bf_secret = bf_secret.encode('ascii')
|
|
|
|
|
|
|
|
|
|
self.s = create_requests_session()
|
|
|
|
|
self.s.headers.update({
|
|
|
|
|
'accept': '*/*',
|
|
|
|
|
'user-agent': 'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/96.0.4664.110 Safari/537.36',
|
|
|
|
|
'content-type': 'text/plain;charset=UTF-8',
|
|
|
|
|
'origin': 'https://www.deezer.com',
|
|
|
|
|
'sec-fetch-site': 'same-origin',
|
|
|
|
|
'sec-fetch-mode': 'same-origin',
|
|
|
|
|
'sec-fetch-dest': 'empty',
|
|
|
|
|
'referer': 'https://www.deezer.com/',
|
|
|
|
|
'accept-language': 'en-US,en;q=0.9',
|
|
|
|
|
})
|
|
|
|
|
|
|
|
|
|
def _api_call(self, method, payload={}):
|
|
|
|
|
api_token = self.api_token if method not in ('deezer.getUserData', 'user.getArl') else ''
|
|
|
|
|
params = {
|
|
|
|
|
'method': method,
|
|
|
|
|
'input': 3,
|
|
|
|
|
'api_version': 1.0,
|
|
|
|
|
'api_token': api_token,
|
2023-11-09 03:14:20 +01:00
|
|
|
'cid': randint(0, 1_000_000_000),
|
2021-12-25 22:38:24 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
resp = self.s.post(self.gw_light_url, params=params, json=payload).json()
|
|
|
|
|
|
|
|
|
|
if resp['error']:
|
2021-12-27 16:48:52 +01:00
|
|
|
type = list(resp['error'].keys())[0]
|
2021-12-25 22:38:24 +01:00
|
|
|
msg = list(resp['error'].values())[0]
|
2021-12-27 16:48:52 +01:00
|
|
|
raise APIError(type, msg, resp['payload'])
|
2021-12-25 22:38:24 +01:00
|
|
|
|
|
|
|
|
if method == 'deezer.getUserData':
|
|
|
|
|
self.api_token = resp['results']['checkForm']
|
|
|
|
|
self.country = resp['results']['COUNTRY']
|
|
|
|
|
self.license_token = resp['results']['USER']['OPTIONS']['license_token']
|
|
|
|
|
self.renew_timestamp = ceil(time())
|
|
|
|
|
self.language = resp['results']['USER']['SETTING']['global']['language']
|
|
|
|
|
|
2021-12-31 00:10:59 +01:00
|
|
|
self.available_formats = ['MP3_128']
|
2021-12-25 22:38:24 +01:00
|
|
|
format_dict = {'web_hq': 'MP3_320', 'web_lossless': 'FLAC'}
|
|
|
|
|
for k, v in format_dict.items():
|
|
|
|
|
if resp['results']['USER']['OPTIONS'][k]:
|
|
|
|
|
self.available_formats.append(v)
|
|
|
|
|
|
|
|
|
|
return resp['results']
|
|
|
|
|
|
|
|
|
|
def login_via_email(self, email, password):
|
2024-02-10 20:33:59 +01:00
|
|
|
# server sends set-cookie header with new sid
|
|
|
|
|
self.s.get('https://www.deezer.com')
|
|
|
|
|
|
2022-02-16 02:02:39 +01:00
|
|
|
password = MD5.new(password.encode()).hexdigest()
|
2021-12-25 22:38:24 +01:00
|
|
|
|
|
|
|
|
params = {
|
|
|
|
|
'app_id': self.client_id,
|
|
|
|
|
'login': email,
|
|
|
|
|
'password': password,
|
2022-02-16 02:02:39 +01:00
|
|
|
'hash': MD5.new((self.client_id + email + password + self.client_secret).encode()).hexdigest(),
|
2021-12-25 22:38:24 +01:00
|
|
|
}
|
|
|
|
|
|
2024-02-10 20:33:59 +01:00
|
|
|
# server sends set-cookie header with account sid
|
2021-12-25 22:38:24 +01:00
|
|
|
json = self.s.get('https://connect.deezer.com/oauth/user_auth.php', params=params).json()
|
|
|
|
|
|
|
|
|
|
if 'error' in json:
|
|
|
|
|
raise self.exception('Error while getting access token, check your credentials')
|
|
|
|
|
|
|
|
|
|
arl = self._api_call('user.getArl')
|
|
|
|
|
|
|
|
|
|
return arl, self.login_via_arl(arl)
|
|
|
|
|
|
|
|
|
|
def login_via_arl(self, arl):
|
|
|
|
|
self.s.cookies.set('arl', arl, domain='.deezer.com')
|
|
|
|
|
user_data = self._api_call('deezer.getUserData')
|
|
|
|
|
|
|
|
|
|
if not user_data['USER']['USER_ID']:
|
|
|
|
|
self.s.cookies.clear()
|
|
|
|
|
raise self.exception('Invalid arl')
|
|
|
|
|
|
|
|
|
|
return user_data
|
|
|
|
|
|
|
|
|
|
def get_track(self, id):
|
|
|
|
|
return self._api_call('deezer.pageTrack', {'sng_id': id})
|
|
|
|
|
|
2021-12-29 21:25:13 +01:00
|
|
|
def get_track_data(self, id):
|
|
|
|
|
return self._api_call('song.getData', {'sng_id': id})
|
|
|
|
|
|
2021-12-25 22:38:24 +01:00
|
|
|
def get_track_lyrics(self, id):
|
|
|
|
|
return self._api_call('song.getLyrics', {'sng_id': id})
|
|
|
|
|
|
|
|
|
|
def get_track_contributors(self, id):
|
|
|
|
|
return self._api_call('song.getData', {'sng_id': id, 'array_default': ['SNG_CONTRIBUTORS']})['SNG_CONTRIBUTORS']
|
|
|
|
|
|
|
|
|
|
def get_track_cover(self, id):
|
|
|
|
|
return self._api_call('song.getData', {'sng_id': id, 'array_default': ['ALB_PICTURE']})['ALB_PICTURE']
|
|
|
|
|
|
|
|
|
|
def get_track_data_by_isrc(self, isrc):
|
|
|
|
|
resp = self.s.get(f'https://api.deezer.com/track/isrc:{isrc}').json()
|
|
|
|
|
if 'error' in resp:
|
|
|
|
|
raise self.exception((resp['error']['type'], resp['error']['message'], resp['error']['code']))
|
2021-12-29 23:47:20 +01:00
|
|
|
|
|
|
|
|
return {
|
|
|
|
|
'SNG_ID': resp['id'],
|
|
|
|
|
'SNG_TITLE': resp['title_short'],
|
|
|
|
|
'VERSION': resp['title_version'],
|
|
|
|
|
'ARTISTS': [{'ART_NAME': a['name']} for a in resp['contributors']],
|
|
|
|
|
'EXPLICIT_LYRICS': str(int(resp['explicit_lyrics'])),
|
|
|
|
|
'ALB_TITLE': resp['album']['title']
|
|
|
|
|
}
|
2021-12-25 22:38:24 +01:00
|
|
|
|
|
|
|
|
def get_album(self, id):
|
2021-12-27 16:48:52 +01:00
|
|
|
try:
|
|
|
|
|
return self._api_call('deezer.pageAlbum', {'alb_id': id, 'lang': self.language})
|
|
|
|
|
except APIError as e:
|
|
|
|
|
if e.payload:
|
|
|
|
|
return self._api_call('deezer.pageAlbum', {'alb_id': e.payload['FALLBACK']['ALB_ID'], 'lang': self.language})
|
|
|
|
|
else:
|
|
|
|
|
raise e
|
2021-12-25 22:38:24 +01:00
|
|
|
|
|
|
|
|
def get_playlist(self, id, nb, start):
|
|
|
|
|
return self._api_call('deezer.pagePlaylist', {'nb': nb, 'start': start, 'playlist_id': id, 'lang': self.language, 'tab': 0, 'tags': True, 'header': True})
|
|
|
|
|
|
|
|
|
|
def get_artist_name(self, id):
|
|
|
|
|
return self._api_call('artist.getData', {'art_id': id, 'array_default': ['ART_NAME']})['ART_NAME']
|
|
|
|
|
|
|
|
|
|
def search(self, query, type, start, nb):
|
|
|
|
|
return self._api_call('search.music', {'query': query, 'start': start, 'nb': nb, 'filter': 'ALL', 'output': type.upper()})
|
|
|
|
|
|
|
|
|
|
def get_artist_album_ids(self, id, start, nb, credited_albums):
|
|
|
|
|
payload = {
|
|
|
|
|
'art_id': id,
|
|
|
|
|
'start': start,
|
|
|
|
|
'nb': nb,
|
|
|
|
|
'filter_role_id': [0,5] if credited_albums else [0],
|
|
|
|
|
'nb_songs': 0,
|
|
|
|
|
'discography_mode': 'all' if credited_albums else None,
|
|
|
|
|
'array_default': ['ALB_ID']
|
|
|
|
|
}
|
|
|
|
|
resp = self._api_call('album.getDiscography', payload)
|
|
|
|
|
return [a['ALB_ID'] for a in resp['data']]
|
|
|
|
|
|
|
|
|
|
def get_track_url(self, id, track_token, track_token_expiry, format):
|
|
|
|
|
# renews license token
|
|
|
|
|
if time() - self.renew_timestamp >= 3600:
|
|
|
|
|
self._api_call('deezer.getUserData')
|
|
|
|
|
|
|
|
|
|
# renews track token
|
|
|
|
|
if time() - track_token_expiry >= 0:
|
|
|
|
|
track_token = self._api_call('song.getData', {'sng_id': id, 'array_default': ['TRACK_TOKEN']})['TRACK_TOKEN']
|
|
|
|
|
|
|
|
|
|
json = {
|
|
|
|
|
'license_token': self.license_token,
|
|
|
|
|
'media': [
|
|
|
|
|
{
|
|
|
|
|
'type': 'FULL',
|
|
|
|
|
'formats': [{'cipher': 'BF_CBC_STRIPE', 'format': format}]
|
|
|
|
|
}
|
|
|
|
|
],
|
|
|
|
|
'track_tokens': [track_token]
|
|
|
|
|
}
|
|
|
|
|
resp = self.s.post('https://media.deezer.com/v1/get_url', json=json).json()
|
|
|
|
|
return resp['data'][0]['media'][0]['sources'][0]['url']
|
|
|
|
|
|
|
|
|
|
def get_legacy_track_url(self, md5_origin, format, id, media_version):
|
2021-12-30 23:54:30 +01:00
|
|
|
format_num = {
|
|
|
|
|
'MP3_MISC': '0',
|
|
|
|
|
'MP3_128': '1',
|
|
|
|
|
'MP4_RA1': '13',
|
|
|
|
|
'MP4_RA2': '14',
|
2022-10-04 15:13:30 +02:00
|
|
|
'MP4_RA3': '15',
|
|
|
|
|
'MHM1_RA1': '16',
|
|
|
|
|
'MHM1_RA2': '17',
|
|
|
|
|
'MHM1_RA3': '18'
|
2021-12-30 23:54:30 +01:00
|
|
|
}[format]
|
|
|
|
|
|
2021-12-25 22:38:24 +01:00
|
|
|
# mashing a bunch of metadata and hashing it with MD5
|
|
|
|
|
info = b"\xa4".join([i.encode() for i in [
|
2021-12-30 23:54:30 +01:00
|
|
|
md5_origin, format_num, str(id), str(media_version)
|
2021-12-25 22:38:24 +01:00
|
|
|
]])
|
|
|
|
|
hash = MD5.new(info).hexdigest()
|
|
|
|
|
|
|
|
|
|
# hash + metadata
|
|
|
|
|
hash_metadata = hash.encode() + b"\xa4" + info + b"\xa4"
|
|
|
|
|
|
|
|
|
|
# padding
|
|
|
|
|
while len(hash_metadata) % 16 > 0:
|
|
|
|
|
hash_metadata += b"\0"
|
|
|
|
|
|
|
|
|
|
# AES encryption
|
|
|
|
|
result = self.legacy_url_cipher.encrypt(hash_metadata).hex()
|
|
|
|
|
|
|
|
|
|
# getting url
|
|
|
|
|
return f"https://cdns-proxy-{md5_origin[0]}.dzcdn.net/mobile/1/{result}"
|
|
|
|
|
|
|
|
|
|
def _get_blowfish_key(self, track_id):
|
|
|
|
|
# yeah, you use the bytes of the hex digest of the hash. bruh moment
|
|
|
|
|
md5_id = MD5.new(str(track_id).encode()).hexdigest().encode('ascii')
|
|
|
|
|
|
|
|
|
|
key = bytes([md5_id[i] ^ md5_id[i + 16] ^ self.bf_secret[i] for i in range(16)])
|
|
|
|
|
|
|
|
|
|
return key
|
|
|
|
|
|
|
|
|
|
def dl_track(self, id, url, path):
|
|
|
|
|
bf_key = self._get_blowfish_key(id)
|
|
|
|
|
req = self.s.get(url, stream=True)
|
|
|
|
|
req.raise_for_status()
|
2021-12-31 15:53:57 +01:00
|
|
|
size = int(req.headers['content-length'])
|
|
|
|
|
bar = tqdm(total=size, unit='B', unit_scale=True)
|
2021-12-25 22:38:24 +01:00
|
|
|
|
|
|
|
|
with open(path, "ab") as file:
|
|
|
|
|
for i, chunk in enumerate(req.iter_content(2048)):
|
|
|
|
|
# every 3rd chunk is encrypted
|
|
|
|
|
if i % 3 == 0 and len(chunk) == 2048:
|
|
|
|
|
# yes, the cipher has to be reset on every chunk.
|
|
|
|
|
# those deezer devs were prob smoking crack when they made this DRM
|
|
|
|
|
cipher = Blowfish.new(bf_key, Blowfish.MODE_CBC, b"\x00\x01\x02\x03\x04\x05\x06\x07")
|
|
|
|
|
chunk = cipher.decrypt(chunk)
|
|
|
|
|
file.write(chunk)
|
2021-12-31 15:53:57 +01:00
|
|
|
bar.update(len(chunk))
|
|
|
|
|
|
|
|
|
|
bar.close()
|
2021-12-30 23:54:30 +01:00
|
|
|
|
|
|
|
|
def check_format(self, md5_origin, format, id, media_version):
|
|
|
|
|
url = self.get_legacy_track_url(md5_origin, format, id, media_version)
|
|
|
|
|
try:
|
2022-09-29 18:09:18 +02:00
|
|
|
resp = self.s.get(url, stream=True)
|
2021-12-30 23:54:30 +01:00
|
|
|
resp.raise_for_status()
|
|
|
|
|
except HTTPError:
|
|
|
|
|
return False
|
|
|
|
|
else:
|
2022-09-29 18:09:18 +02:00
|
|
|
return True
|