summaryrefslogtreecommitdiffstats
path: root/module/plugins/hoster/ShareonlineBiz.py
diff options
context:
space:
mode:
Diffstat (limited to 'module/plugins/hoster/ShareonlineBiz.py')
-rw-r--r--module/plugins/hoster/ShareonlineBiz.py128
1 files changed, 128 insertions, 0 deletions
diff --git a/module/plugins/hoster/ShareonlineBiz.py b/module/plugins/hoster/ShareonlineBiz.py
new file mode 100644
index 000000000..bc1951602
--- /dev/null
+++ b/module/plugins/hoster/ShareonlineBiz.py
@@ -0,0 +1,128 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+
+import os
+import os.path
+import re
+import tempfile
+from time import time
+from base64 import b64decode
+import hashlib
+import random
+from time import sleep
+
+from module.plugins.Hoster import Hoster
+from module.network.Request import getURL
+from module.plugins.Plugin import chunks
+
+
+def getInfo(urls):
+ api_url_base = "http://www.share-online.biz/linkcheck/linkcheck.php"
+
+ for chunk in chunks(urls, 90):
+ api_param_file = {"links": "\n".join(x.replace("http://www.share-online.biz/dl/","") for x in chunk)} #api only supports old style links
+ src = getURL(api_url_base, post=api_param_file)
+ result = []
+ for i, res in enumerate(src.split("\n")):
+ if not res:
+ continue
+ fields = res.split(";")
+
+ if fields[1] == "OK":
+ status = 2
+ elif fields[1] in ("DELETED", "NOT FOUND"):
+ status = 1
+ else:
+ status = 3
+
+ result.append((fields[2], int(fields[3]), status, chunk[i]))
+ yield result
+
+class ShareonlineBiz(Hoster):
+ __name__ = "ShareonlineBiz"
+ __type__ = "hoster"
+ __pattern__ = r"(?:http://)?(?:www.)?share-online.biz/(download.php\?id=|dl/)"
+ __version__ = "0.2"
+ __description__ = """Shareonline.biz Download Hoster"""
+ __author_name__ = ("spoob", "mkaay")
+ __author_mail__ = ("spoob@pyload.org", "mkaay@mkaay.de")
+
+ def setup(self):
+ #self.req.canContinue = self.multiDL = True if self.account else False
+ # range request not working?
+ self.multiDL = True if self.account else False
+
+ def process(self, pyfile):
+ self.convertURL()
+ self.downloadAPIData()
+ pyfile.name = self.api_data["filename"]
+ pyfile.sync()
+
+ self.downloadHTML()
+
+ self.download(self.getFileUrl(), cookies=True)
+
+ def downloadAPIData(self):
+ api_url_base = "http://www.share-online.biz/linkcheck/linkcheck.php?md5=1"
+ api_param_file = {"links": self.pyfile.url.replace("http://www.share-online.biz/dl/","")} #api only supports old style links
+ src = self.load(api_url_base, cookies=False, post=api_param_file)
+
+ fields = src.split(";")
+ self.api_data = {}
+ self.api_data["fileid"] = fields[0]
+ self.api_data["status"] = fields[1]
+ if not self.api_data["status"] == "OK":
+ self.offline()
+ self.api_data["filename"] = fields[2]
+ self.api_data["size"] = fields[3] # in bytes
+ self.api_data["checksum"] = fields[4].strip().lower().replace("\n\n", "") # md5
+
+ def downloadHTML(self):
+ self.html = self.load(self.pyfile.url, cookies=True)
+
+ if not self.account:
+ html = self.load("%s/free/" % self.pyfile.url, post={"dl_free":"1"}, cookies=True)
+ if re.search(r"/failure/full/1", self.req.lastEffectiveURL):
+ self.setWait(120)
+ self.log.debug("%s: no free slots, waiting 120 seconds" % (self.__name__))
+ self.wait()
+ self.retry()
+ captcha = self.decryptCaptcha("http://www.share-online.biz/captcha.php", get={"rand":"0.%s" % random.randint(10**15,10**16)}, cookies=True)
+
+ self.log.debug("%s Captcha: %s" % (self.__name__, captcha))
+ sleep(3)
+
+ html = self.load(self.pyfile.url, post={"captchacode": captcha}, cookies=True)
+ if re.search(r"Der Download ist Ihnen zu langsam", html):
+ #m = re.search("var timeout='(\d+)';", self.html[1])
+ #self.waitUntil = time() + int(m.group(1)) if m else 30
+ return True
+
+ self.retry()
+ else:
+ return True
+
+ def convertURL(self):
+ self.pyfile.url = self.pyfile.url.replace("http://www.share-online.biz/download.php?id=", "http://www.share-online.biz/dl/")
+
+ def getFileUrl(self):
+ """ returns the absolute downloadable filepath
+ """
+ if self.account:
+ return b64decode(re.search('var dl="(.*?)"', self.html).group(1))
+ file_url_pattern = 'loadfilelink\.decode\("([^"]+)'
+ return b64decode(re.search(file_url_pattern, self.html).group(1))
+
+ def checksum(self, local_file):
+ if self.api_data and self.api_data["checksum"]:
+ h = hashlib.md5()
+ f = open(local_file, "rb")
+ h.update(f.read())
+ f.close()
+ hexd = h.hexdigest()
+ if hexd == self.api_data["checksum"]:
+ return (True, 0)
+ else:
+ return (False, 1)
+ else:
+ return (True, 5)