diff options
Diffstat (limited to 'module/plugins/crypter/RelinkUs.py')
-rw-r--r-- | module/plugins/crypter/RelinkUs.py | 263 |
1 files changed, 0 insertions, 263 deletions
diff --git a/module/plugins/crypter/RelinkUs.py b/module/plugins/crypter/RelinkUs.py deleted file mode 100644 index 1b120fc77..000000000 --- a/module/plugins/crypter/RelinkUs.py +++ /dev/null @@ -1,263 +0,0 @@ -# -*- coding: utf-8 -*- - -import base64 -import binascii -import re -import os - -from Crypto.Cipher import AES -from module.plugins.Crypter import Crypter - - -class RelinkUs(Crypter): - __name__ = "RelinkUs" - __type__ = "crypter" - __version__ = "3.0" - - __pattern__ = r'http://(?:www\.)?relink.us/(f/|((view|go).php\?id=))(?P<id>.+)' - - __description__ = """Relink.us decrypter plugin""" - __author_name__ = "fragonib" - __author_mail__ = "fragonib[AT]yahoo[DOT]es" - - # Constants - PREFERRED_LINK_SOURCES = ["cnl2", "dlc", "web"] - - OFFLINE_TOKEN = r'<title>Tattooside' - PASSWORD_TOKEN = r'container_password\.php' - PASSWORD_ERROR_ROKEN = r'You have entered an incorrect password' - PASSWORD_SUBMIT_URL = r'http://www\.relink\.us/container_password\.php' - CAPTCHA_TOKEN = r'container_captcha\.php' - CAPTCHA_ERROR_ROKEN = r'You have solved the captcha wrong' - CAPTCHA_IMG_URL = r'http://www\.relink\.us/core/captcha/circlecaptcha\.php' - CAPTCHA_SUBMIT_URL = r'http://www\.relink\.us/container_captcha\.php' - FILE_TITLE_REGEX = r'<th>Title</th><td><i>(.*)</i></td></tr>' - FILE_NOTITLE = r'No title' - - CNL2_FORM_REGEX = r'<form id="cnl_form-(.*?)</form>' - CNL2_FORMINPUT_REGEX = r'<input.*?name="%s".*?value="(.*?)"' - CNL2_JK_KEY = "jk" - CNL2_CRYPTED_KEY = "crypted" - DLC_LINK_REGEX = r'<a href=".*?" class="dlc_button" target="_blank">' - DLC_DOWNLOAD_URL = r'http://www\.relink\.us/download\.php' - WEB_FORWARD_REGEX = r"getFile\('(?P<link>.+)'\)" - WEB_FORWARD_URL = r'http://www\.relink\.us/frame\.php' - WEB_LINK_REGEX = r'<iframe name="Container" height="100%" frameborder="no" width="100%" src="(?P<link>.+)"></iframe>' - - - def setup(self): - self.fileid = None - self.package = None - self.password = None - self.html = None - self.captcha = False - - def decrypt(self, pyfile): - # Init - self.initPackage(pyfile) - - # Request package - self.requestPackage() - - # Check for online - if not self.isOnline(): - self.offline() - - # Check for protection - if self.isPasswordProtected(): - self.unlockPasswordProtection() - self.handleErrors() - - if self.isCaptchaProtected(): - self.captcha = True - self.unlockCaptchaProtection() - self.handleErrors() - - # Get package name and folder - (package_name, folder_name) = self.getPackageInfo() - - # Extract package links - package_links = [] - for sources in self.PREFERRED_LINK_SOURCES: - package_links.extend(self.handleLinkSource(sources)) - if package_links: # use only first source which provides links - break - package_links = set(package_links) - - # Pack - if package_links: - self.packages = [(package_name, package_links, folder_name)] - else: - self.fail('Could not extract any links') - - def initPackage(self, pyfile): - self.fileid = re.match(self.__pattern__, pyfile.url).group('id') - self.package = pyfile.package() - self.password = self.getPassword() - - def requestPackage(self): - self.html = self.load(self.pyfile.url, decode=True) - - def isOnline(self): - if self.OFFLINE_TOKEN in self.html: - self.logDebug("File not found") - return False - return True - - def isPasswordProtected(self): - if self.PASSWORD_TOKEN in self.html: - self.logDebug("Links are password protected") - return True - - def isCaptchaProtected(self): - if self.CAPTCHA_TOKEN in self.html: - self.logDebug("Links are captcha protected") - return True - return False - - def unlockPasswordProtection(self): - self.logDebug("Submitting password [%s] for protected links" % self.password) - passwd_url = self.PASSWORD_SUBMIT_URL + "?id=%s" % self.fileid - passwd_data = {'id': self.fileid, 'password': self.password, 'pw': 'submit'} - self.html = self.load(passwd_url, post=passwd_data, decode=True) - - def unlockCaptchaProtection(self): - self.logDebug("Request user positional captcha resolving") - captcha_img_url = self.CAPTCHA_IMG_URL + "?id=%s" % self.fileid - coords = self.decryptCaptcha(captcha_img_url, forceUser=True, imgtype="png", result_type='positional') - self.logDebug("Captcha resolved, coords [%s]" % str(coords)) - captcha_post_url = self.CAPTCHA_SUBMIT_URL + "?id=%s" % self.fileid - captcha_post_data = {'button.x': coords[0], 'button.y': coords[1], 'captcha': 'submit'} - self.html = self.load(captcha_post_url, post=captcha_post_data, decode=True) - - def getPackageInfo(self): - name = folder = None - - # Try to get info from web - m = re.search(self.FILE_TITLE_REGEX, self.html) - if m is not None: - title = m.group(1).strip() - if not self.FILE_NOTITLE in title: - name = folder = title - self.logDebug("Found name [%s] and folder [%s] in package info" % (name, folder)) - - # Fallback to defaults - if not name or not folder: - name = self.package.name - folder = self.package.folder - self.logDebug("Package info not found, defaulting to pyfile name [%s] and folder [%s]" % (name, folder)) - - # Return package info - return name, folder - - def handleErrors(self): - if self.PASSWORD_ERROR_ROKEN in self.html: - msg = "Incorrect password, please set right password on 'Edit package' form and retry" - self.logDebug(msg) - self.fail(msg) - - if self.captcha: - if self.CAPTCHA_ERROR_ROKEN in self.html: - self.logDebug("Invalid captcha, retrying") - self.invalidCaptcha() - self.retry() - else: - self.correctCaptcha() - - def handleLinkSource(self, source): - if source == 'cnl2': - return self.handleCNL2Links() - elif source == 'dlc': - return self.handleDLCLinks() - elif source == 'web': - return self.handleWEBLinks() - else: - self.fail('Unknown source [%s] (this is probably a bug)' % source) - - def handleCNL2Links(self): - self.logDebug("Search for CNL2 links") - package_links = [] - m = re.search(self.CNL2_FORM_REGEX, self.html, re.DOTALL) - if m is not None: - cnl2_form = m.group(1) - try: - (vcrypted, vjk) = self._getCipherParams(cnl2_form) - for (crypted, jk) in zip(vcrypted, vjk): - package_links.extend(self._getLinks(crypted, jk)) - except: - self.logDebug("Unable to decrypt CNL2 links") - return package_links - - def handleDLCLinks(self): - self.logDebug('Search for DLC links') - package_links = [] - m = re.search(self.DLC_LINK_REGEX, self.html) - if m is not None: - container_url = self.DLC_DOWNLOAD_URL + "?id=%s&dlc=1" % self.fileid - self.logDebug("Downloading DLC container link [%s]" % container_url) - try: - dlc = self.load(container_url) - dlc_filename = self.fileid + ".dlc" - dlc_filepath = os.path.join(self.config['general']['download_folder'], dlc_filename) - f = open(dlc_filepath, "wb") - f.write(dlc) - f.close() - package_links.append(dlc_filepath) - except: - self.logDebug("Unable to download DLC container") - return package_links - - def handleWEBLinks(self): - self.logDebug("Search for WEB links") - package_links = [] - fw_params = re.findall(self.WEB_FORWARD_REGEX, self.html) - self.logDebug("Decrypting %d Web links" % len(fw_params)) - for index, fw_param in enumerate(fw_params): - try: - fw_url = self.WEB_FORWARD_URL + "?%s" % fw_param - self.logDebug("Decrypting Web link %d, %s" % (index + 1, fw_url)) - fw_response = self.load(fw_url, decode=True) - dl_link = re.search(self.WEB_LINK_REGEX, fw_response).group('link') - package_links.append(dl_link) - except Exception, detail: - self.logDebug("Error decrypting Web link %s, %s" % (index, detail)) - self.setWait(4) - self.wait() - return package_links - - def _getCipherParams(self, cnl2_form): - # Get jk - jk_re = self.CNL2_FORMINPUT_REGEX % self.CNL2_JK_KEY - vjk = re.findall(jk_re, cnl2_form, re.IGNORECASE) - - # Get crypted - crypted_re = self.CNL2_FORMINPUT_REGEX % RelinkUs.CNL2_CRYPTED_KEY - vcrypted = re.findall(crypted_re, cnl2_form, re.IGNORECASE) - - # Log and return - self.logDebug("Detected %d crypted blocks" % len(vcrypted)) - return vcrypted, vjk - - def _getLinks(self, crypted, jk): - # Get key - jreturn = self.js.eval("%s f()" % jk) - self.logDebug("JsEngine returns value [%s]" % jreturn) - key = binascii.unhexlify(jreturn) - - # Decode crypted - crypted = base64.standard_b64decode(crypted) - - # Decrypt - Key = key - IV = key - obj = AES.new(Key, AES.MODE_CBC, IV) - text = obj.decrypt(crypted) - - # Extract links - text = text.replace("\x00", "").replace("\r", "") - links = text.split("\n") - links = filter(lambda x: x != "", links) - - # Log and return - self.logDebug("Package has %d links" % len(links)) - return links |