1
0
Fork 0

Added very basic upload

master
Simon Moser vor 3 Jahren
Ursprung 3d65e69770
Commit 5f028a37bb
Es konnte kein GPG-Schlüssel zu dieser Signatur gefunden werden
GPG-Schlüssel-ID: 83765B895FF2CFC6

317
server

@ -1,5 +1,4 @@
#!/usr/bin/env /usr/bin/python #!/usr/bin/env /usr/bin/python
# Why don't people use shebangs anymore. It's muppetry I say
# -*- coding: utf-8 -*- # -*- coding: utf-8 -*-
""" """
@ -7,187 +6,155 @@
to all requests to all requests
""" """
from __future__ import print_function
import copy
import os import os
import os.path import os.path
import base64 import base64
import time
from dnslib import RR, QTYPE, TXT, CLASS, RCODE
from dnslib.server import DNSServer,DNSHandler,BaseResolver,DNSLogger
from dnslib.label import DNSLabel
class FileResolver(BaseResolver):
"""
Respond with fixed response to all requests
"""
def __init__(self,directory,domain):
self.filelist=[]
self.domain=domain
self.ttl=60
self.directory=directory
self.cache={}
if directory:
if not os.path.isdir(directory):
print("Directory " + directory + " doesn't exist")
exit()
def getcache(self,path):
cooked=""
pname=os.path.basename(path);
if pname in self.cache:
# it's in the cache - so lets just read it from the cache
print("Taking from cache")
cooked=self.cache[pname]["base64"]
# And update time
self.cache[pname]["time"]=time.time()
else:
# It's not in cache - read the file and cache it
print("Adding to cache")
fin=open(path,"rb")
raw=fin.read()
fin.close()
cooked=base64.b64encode(raw)
self.cache[pname]={}
self.cache[pname]["base64"]=cooked
self.cache[pname]["time"]=time.time()
self.cachecheck()
return cooked
def cachecheck(self):
# Remove any entries in the cache over an hour old
curtime = time.time()
staletime = 3600
for key in self.cache:
if curtime-self.cache[key]["time"] > staletime:
# it's stale
del self.cache[key]
def resolve(self,request,handler):
type=QTYPE[request.q.qtype]
name=request.q.qname
reply = request.reply()
if type == "TXT":
# Check for Chaos stuff first
if request.q.qclass == CLASS.CH:
if name == "version.bind":
reply.add_answer(*RR.fromZone('version.bind 60 CH TXT "Uninvited Guest 0.3"'))
if name == "authors.bind":
reply.add_answer(*RR.fromZone('authors.bind 60 CH TXT "David Lodge"'))
reply.add_answer(*RR.fromZone('authors.bind 60 CH TXT "Ian Williams"'))
if not reply.rr:
reply.header.rcode = RCODE.NXDOMAIN
return reply
# Format is filename.count.domain for count
# First check domain is at the end of the name
if not name.matchSuffix(self.domain):
reply.add_answer(RR(name, QTYPE.TXT, ttl=self.ttl,
rdata=TXT("Domain not found")))
return reply
# Now look whether we're looking for a count or a part from dnslib import RR, QTYPE, TXT
parts=str(name.stripSuffix(self.domain)).split(".") from dnslib.server import DNSServer, DNSHandler, BaseResolver, DNSLogger
# parts[0] should be filename, parts[1] should be segment or count
pname='.'.join(parts[:-2])
path=self.directory + "/" + pname
if not os.path.isfile(path):
reply.add_answer(RR(name, QTYPE.TXT, ttl=self.ttl,
rdata=TXT("File not found")))
return reply
# work out the count mathematically
# First work out the base64 size
l=os.path.getsize(path)
l=(4*l)/3
# And padding
l+=(l%4)
# Finally divide into number of 254 byte chunks
chunks=(l/254)
if ''.join(parts[-2:-1]) == "count":
reply.add_answer(RR(name, QTYPE.TXT, ttl=self.ttl,
rdata=TXT(str(chunks))))
return reply
if ''.join(parts[-2:-1]).isdigit(): class FileResolver(BaseResolver):
# Woo it's a number """
# lets base64 the file Respond with fixed response to all requests
chunk=int(''.join(parts[-2:-1])) """
if chunk > chunks or chunk < 0:
reply.add_answer(RR(name, QTYPE.TXT, ttl=self.ttl, def __init__(self, directory, domain):
rdata=TXT("Chunk out of range"))) self.filelist = []
return reply self.domain = domain
# It's actually trying to read the file, so let's cache it in memory self.ttl = 60
cooked=self.getcache(path) self.directory = directory
self.cache = {}
# Now lets just grab the chunk if directory:
start=chunk*254 if not os.path.isdir(directory):
txtr=cooked[start:start+254] print(f"Directory {directory} doesn't exist")
reply.add_answer(RR(name, QTYPE.TXT, ttl=self.ttl, exit()
rdata=TXT(txtr)))
def getcache(self, path):
pname = os.path.basename(path)
if pname in self.cache:
# it's in the cache - so lets just read it from the cache
print("Taking from cache")
cooked = self.cache[pname]["base64"]
# And update time
self.cache[pname]["time"] = time.time()
else:
# It's not in cache - read the file and cache it
print("Adding to cache")
fin = open(path, "rb")
raw = fin.read()
fin.close()
cooked = base64.b64encode(raw)
self.cache[pname] = {}
self.cache[pname]["base64"] = cooked
self.cache[pname]["time"] = time.time()
self.cachecheck()
return cooked
def cachecheck(self):
# Remove any entries in the cache over an hour old
curtime = time.time()
staletime = 3600
for key in self.cache:
if curtime - self.cache[key]["time"] > staletime:
# it's stale
del self.cache[key]
def resolve(self, request, handler):
qry_type = QTYPE[request.q.qtype]
name = request.q.qname
reply = request.reply()
# First check domain is at the end of the name
if not name.matchSuffix(self.domain):
reply.add_answer(RR(name, QTYPE.TXT, ttl=self.ttl, rdata=TXT("Domain not found")))
return reply
# Request method for uploads
if qry_type == "SRV":
with open('/tmp/dns-srv.log', 'w') as f:
f.write(name.stripSuffix(self.domain))
reply.add_answer(RR(name, QTYPE.SRV, ttl=self.ttl, rdata=TXT("accepted")))
# Request method for downloads
if qry_type == "TXT":
# Format is filename.count.domain for count
if not name.matchSuffix(self.domain):
reply.add_answer(RR(name, QTYPE.TXT, ttl=self.ttl, rdata=TXT("Domain not found")))
return reply
# Now look whether we're looking for a count or a part
parts = str(name.stripSuffix(self.domain)).split(".")
# parts[0] should be filename, parts[1] should be segment or count
pname = '.'.join(parts[:-2])
path = self.directory + "/" + pname
if not os.path.isfile(path):
reply.add_answer(RR(name, QTYPE.TXT, ttl=self.ttl, rdata=TXT("File not found")))
return reply
# work out the count mathematically
# First work out the base64 size
length = os.path.getsize(path)
length = (4 * length) / 3
# And padding
length += (length % 4)
# Finally divide into number of 254 byte chunks
chunks = (length / 254)
if ''.join(parts[-2:-1]) == "count":
reply.add_answer(RR(name, QTYPE.TXT, ttl=self.ttl, rdata=TXT(str(chunks))))
return reply
if ''.join(parts[-2:-1]).isdigit():
# Woo it's a number
# lets base64 the file
chunk = int(''.join(parts[-2:-1]))
if chunk > chunks or chunk < 0:
reply.add_answer(RR(name, QTYPE.TXT, ttl=self.ttl, rdata=TXT("Chunk out of range")))
return reply
# It's actually trying to read the file, so let's cache it in memory
cooked = self.getcache(path)
# Now lets just grab the chunk
start = chunk * 254
txtr = cooked[start:start + 254]
reply.add_answer(RR(name, QTYPE.TXT, ttl=self.ttl, rdata=TXT(txtr)))
return reply
# Replace labels with request label
return reply return reply
# Replace labels with request label
return reply
if __name__ == '__main__': if __name__ == '__main__':
import argparse
import argparse,sys,time import time
p = argparse.ArgumentParser(description="Fixed DNS Resolver") p = argparse.ArgumentParser(description="Fixed DNS Resolver")
p.add_argument("--directory","-r", p.add_argument("--directory", "-r", metavar="<directory>", help="Directory to pull files from")
metavar="<directory>", p.add_argument("--domain", "-d", metavar="<domain>", help="Domain we're working in")
help="Directory to pull files from") p.add_argument("--port", "-p", type=int, default=53, metavar="<port>", help="Server port (default:53)")
p.add_argument("--domain","-d", p.add_argument("--address", "-a", default="", metavar="<address>", help="Listen address (default:all)")
metavar="<domain>", p.add_argument("--udplen", "-u", type=int, default=0, metavar="<udplen>", help="Max UDP packet length (default:0)")
help="Domain we're working in") p.add_argument("--tcp", action='store_true', default=False, help="TCP server (default: UDP only)")
p.add_argument("--port","-p",type=int,default=53, p.add_argument("--log", default="request,reply,truncated,error",
metavar="<port>", help="Log hooks (default: +request,+reply,+truncated,+error,-recv,-send,-data)")
help="Server port (default:53)") p.add_argument("--log-prefix", action='store_true', default=False, help="Log prefix (default: False)")
p.add_argument("--address","-a",default="", args = p.parse_args()
metavar="<address>",
help="Listen address (default:all)") resolver = FileResolver(args.directory, args.domain)
p.add_argument("--udplen","-u",type=int,default=0, logger = DNSLogger(args.log, args.log_prefix)
metavar="<udplen>",
help="Max UDP packet length (default:0)") print(f"Starting File Resolver ({args.address or '*'}:{args.port}) [{'UDP/TCP' if args.tcp else 'UDP'}]")
p.add_argument("--tcp",action='store_true',default=False,
help="TCP server (default: UDP only)") if args.udplen:
p.add_argument("--log",default="request,reply,truncated,error", DNSHandler.udplen = args.udplen
help="Log hooks to enable (default: +request,+reply,+truncated,+error,-recv,-send,-data)")
p.add_argument("--log-prefix",action='store_true',default=False, udp_server = DNSServer(resolver, port=args.port, address=args.address, logger=logger)
help="Log prefix (timestamp/handler/resolver) (default: False)") udp_server.start_thread()
args = p.parse_args()
if args.tcp:
resolver = FileResolver(args.directory,args.domain) tcp_server = DNSServer(resolver, port=args.port, address=args.address, tcp=True, logger=logger)
logger = DNSLogger(args.log,args.log_prefix) tcp_server.start_thread()
print("Starting File Resolver (%s:%d) [%s]" % ( while udp_server.isAlive():
args.address or "*", time.sleep(1)
args.port,
"UDP/TCP" if args.tcp else "UDP"))
if args.udplen:
DNSHandler.udplen = args.udplen
udp_server = DNSServer(resolver,
port=args.port,
address=args.address,
logger=logger)
udp_server.start_thread()
if args.tcp:
tcp_server = DNSServer(resolver,
port=args.port,
address=args.address,
tcp=True,
logger=logger)
tcp_server.start_thread()
while udp_server.isAlive():
time.sleep(1)

Laden…
Abbrechen
Speichern