sshuttle/firewall.py

461 lines
15 KiB
Python
Raw Permalink Normal View History

import re, errno, socket, select, struct
import compat.ssubprocess as ssubprocess
import helpers, ssyslog
from helpers import *
# python doesn't have a definition for this
IPPROTO_DIVERT = 254
def nonfatal(func, *args):
try:
func(*args)
except Fatal, e:
log('error: %s\n' % e)
def ipt_chain_exists(name):
argv = ['iptables', '-t', 'nat', '-nL']
p = ssubprocess.Popen(argv, stdout = ssubprocess.PIPE)
for line in p.stdout:
if line.startswith('Chain %s ' % name):
return True
rv = p.wait()
if rv:
raise Fatal('%r returned %d' % (argv, rv))
def ipt(*args):
argv = ['iptables', '-t', 'nat'] + list(args)
debug1('>> %s\n' % ' '.join(argv))
rv = ssubprocess.call(argv)
if rv:
raise Fatal('%r returned %d' % (argv, rv))
_no_ttl_module = False
def ipt_ttl(*args):
global _no_ttl_module
if not _no_ttl_module:
# we avoid infinite loops by generating server-side connections
# with ttl 42. This makes the client side not recapture those
# connections, in case client == server.
try:
argsplus = list(args) + ['-m', 'ttl', '!', '--ttl', '42']
ipt(*argsplus)
except Fatal:
ipt(*args)
# we only get here if the non-ttl attempt succeeds
log('sshuttle: warning: your iptables is missing '
'the ttl module.\n')
_no_ttl_module = True
else:
ipt(*args)
# We name the chain based on the transproxy port number so that it's possible
# to run multiple copies of sshuttle at the same time. Of course, the
# multiple copies shouldn't have overlapping subnets, or only the most-
# recently-started one will win (because we use "-I OUTPUT 1" instead of
# "-A OUTPUT").
def do_iptables(port, dnsport, subnets):
chain = 'sshuttle-%s' % port
# basic cleanup/setup of chains
if ipt_chain_exists(chain):
nonfatal(ipt, '-D', 'OUTPUT', '-j', chain)
nonfatal(ipt, '-D', 'PREROUTING', '-j', chain)
nonfatal(ipt, '-F', chain)
ipt('-X', chain)
if subnets or dnsport:
ipt('-N', chain)
ipt('-F', chain)
ipt('-I', 'OUTPUT', '1', '-j', chain)
ipt('-I', 'PREROUTING', '1', '-j', chain)
if subnets:
# create new subnet entries. Note that we're sorting in a very
# particular order: we need to go from most-specific (largest swidth)
# to least-specific, and at any given level of specificity, we want
# excludes to come first. That's why the columns are in such a non-
# intuitive order.
for swidth,sexclude,snet in sorted(subnets, reverse=True):
if sexclude:
ipt('-A', chain, '-j', 'RETURN',
'--dest', '%s/%s' % (snet,swidth),
'-p', 'tcp')
else:
ipt_ttl('-A', chain, '-j', 'REDIRECT',
'--dest', '%s/%s' % (snet,swidth),
'-p', 'tcp',
'--to-ports', str(port))
if dnsport:
nslist = resolvconf_nameservers()
for ip in nslist:
ipt_ttl('-A', chain, '-j', 'REDIRECT',
'--dest', '%s/32' % ip,
'-p', 'udp',
'--dport', '53',
'--to-ports', str(dnsport))
def ipfw_rule_exists(n):
argv = ['ipfw', 'list']
p = ssubprocess.Popen(argv, stdout = ssubprocess.PIPE)
found = False
for line in p.stdout:
if line.startswith('%05d ' % n):
if not ('ipttl 42' in line
or ('skipto %d' % (n+1)) in line
or 'check-state' in line):
log('non-sshuttle ipfw rule: %r\n' % line.strip())
raise Fatal('non-sshuttle ipfw rule #%d already exists!' % n)
found = True
rv = p.wait()
if rv:
raise Fatal('%r returned %d' % (argv, rv))
return found
_oldctls = {}
def _fill_oldctls(prefix):
argv = ['sysctl', prefix]
p = ssubprocess.Popen(argv, stdout = ssubprocess.PIPE)
for line in p.stdout:
assert(line[-1] == '\n')
(k,v) = line[:-1].split(': ', 1)
_oldctls[k] = v
rv = p.wait()
if rv:
raise Fatal('%r returned %d' % (argv, rv))
if not line:
raise Fatal('%r returned no data' % (argv,))
def _sysctl_set(name, val):
argv = ['sysctl', '-w', '%s=%s' % (name, val)]
debug1('>> %s\n' % ' '.join(argv))
firewall.py: MacOS: permanently set the net.inet.ip.scopedroute sysctl. If this sysctl isn't set to 0 at the time your network interface is brought up, and we later change it, then the MacOS (10.6.6 at least) ARP table gets totally confused and networking stops working about 15 minutes later, until you down and re-up the interface. The symptom is that pings outside your LAN would give results like this: ping: sendto: no route to host and "arp -a -n" would show *two* entries for your default gateway instead of just one. sshuttle was helpfully putting the sysctl back the way it was when it shuts down, so you would fix your network by downing the interface, so sshuttle would abort and change the sysctl back, then you would re-up the interface, then restart sshuttle, and sshuttle would change the sysctl back and restart the cycle: it would break again a few minutes later. That's annoying, and it gives sshuttle a bad reputation for being the thing that breaks your network. I can't find a *really* good workaround for the bug, so barring that, let's just permanently set the sysctl to 0 and not change it back on exit. That should just leave your computer back how it worked in MacOS 10.5, as far as I know, which seems harmless. At least I've been running my Mac that way for a few days and I haven't seen any weirdness. Now, doing *that* would still mean that the first sshuttle session after a reboot would still break the network, since sysctl changes are lost on reboot. Thus, let's be extra hardcore and write it to /etc/sysctl.conf so that it goes the way we want it after a reboot. Thus, sshuttle should break your network at most once. Which still sucks, but hopefully nobody will notice.
2011-02-05 06:55:40 +01:00
return ssubprocess.call(argv, stdout = open('/dev/null', 'w'))
_changedctls = []
firewall.py: MacOS: permanently set the net.inet.ip.scopedroute sysctl. If this sysctl isn't set to 0 at the time your network interface is brought up, and we later change it, then the MacOS (10.6.6 at least) ARP table gets totally confused and networking stops working about 15 minutes later, until you down and re-up the interface. The symptom is that pings outside your LAN would give results like this: ping: sendto: no route to host and "arp -a -n" would show *two* entries for your default gateway instead of just one. sshuttle was helpfully putting the sysctl back the way it was when it shuts down, so you would fix your network by downing the interface, so sshuttle would abort and change the sysctl back, then you would re-up the interface, then restart sshuttle, and sshuttle would change the sysctl back and restart the cycle: it would break again a few minutes later. That's annoying, and it gives sshuttle a bad reputation for being the thing that breaks your network. I can't find a *really* good workaround for the bug, so barring that, let's just permanently set the sysctl to 0 and not change it back on exit. That should just leave your computer back how it worked in MacOS 10.5, as far as I know, which seems harmless. At least I've been running my Mac that way for a few days and I haven't seen any weirdness. Now, doing *that* would still mean that the first sshuttle session after a reboot would still break the network, since sysctl changes are lost on reboot. Thus, let's be extra hardcore and write it to /etc/sysctl.conf so that it goes the way we want it after a reboot. Thus, sshuttle should break your network at most once. Which still sucks, but hopefully nobody will notice.
2011-02-05 06:55:40 +01:00
def sysctl_set(name, val, permanent=False):
PREFIX = 'net.inet.ip'
assert(name.startswith(PREFIX + '.'))
val = str(val)
if not _oldctls:
_fill_oldctls(PREFIX)
if not (name in _oldctls):
debug1('>> No such sysctl: %r\n' % name)
return False
oldval = _oldctls[name]
if val != oldval:
firewall.py: MacOS: permanently set the net.inet.ip.scopedroute sysctl. If this sysctl isn't set to 0 at the time your network interface is brought up, and we later change it, then the MacOS (10.6.6 at least) ARP table gets totally confused and networking stops working about 15 minutes later, until you down and re-up the interface. The symptom is that pings outside your LAN would give results like this: ping: sendto: no route to host and "arp -a -n" would show *two* entries for your default gateway instead of just one. sshuttle was helpfully putting the sysctl back the way it was when it shuts down, so you would fix your network by downing the interface, so sshuttle would abort and change the sysctl back, then you would re-up the interface, then restart sshuttle, and sshuttle would change the sysctl back and restart the cycle: it would break again a few minutes later. That's annoying, and it gives sshuttle a bad reputation for being the thing that breaks your network. I can't find a *really* good workaround for the bug, so barring that, let's just permanently set the sysctl to 0 and not change it back on exit. That should just leave your computer back how it worked in MacOS 10.5, as far as I know, which seems harmless. At least I've been running my Mac that way for a few days and I haven't seen any weirdness. Now, doing *that* would still mean that the first sshuttle session after a reboot would still break the network, since sysctl changes are lost on reboot. Thus, let's be extra hardcore and write it to /etc/sysctl.conf so that it goes the way we want it after a reboot. Thus, sshuttle should break your network at most once. Which still sucks, but hopefully nobody will notice.
2011-02-05 06:55:40 +01:00
rv = _sysctl_set(name, val)
if rv==0 and permanent:
debug1('>> ...saving permanently in /etc/sysctl.conf\n')
f = open('/etc/sysctl.conf', 'a')
f.write('\n'
'# Added by sshuttle\n'
'%s=%s\n' % (name, val))
f.close()
else:
_changedctls.append(name)
return True
def _udp_unpack(p):
src = (socket.inet_ntoa(p[12:16]), struct.unpack('!H', p[20:22])[0])
dst = (socket.inet_ntoa(p[16:20]), struct.unpack('!H', p[22:24])[0])
return src, dst
def _udp_repack(p, src, dst):
addrs = socket.inet_aton(src[0]) + socket.inet_aton(dst[0])
ports = struct.pack('!HH', src[1], dst[1])
return p[:12] + addrs + ports + p[24:]
_real_dns_server = [None]
def _handle_diversion(divertsock, dnsport):
p,tag = divertsock.recvfrom(4096)
src,dst = _udp_unpack(p)
debug3('got diverted packet from %r to %r\n' % (src, dst))
if dst[1] == 53:
# outgoing DNS
debug3('...packet is a DNS request.\n')
_real_dns_server[0] = dst
dst = ('127.0.0.1', dnsport)
elif src[1] == dnsport:
if islocal(src[0]):
debug3('...packet is a DNS response.\n')
src = _real_dns_server[0]
else:
log('weird?! unexpected divert from %r to %r\n' % (src, dst))
assert(0)
newp = _udp_repack(p, src, dst)
divertsock.sendto(newp, tag)
def ipfw(*args):
argv = ['ipfw', '-q'] + list(args)
debug1('>> %s\n' % ' '.join(argv))
rv = ssubprocess.call(argv)
if rv:
raise Fatal('%r returned %d' % (argv, rv))
def do_ipfw(port, dnsport, subnets):
sport = str(port)
xsport = str(port+1)
# cleanup any existing rules
if ipfw_rule_exists(port):
ipfw('delete', sport)
while _changedctls:
name = _changedctls.pop()
oldval = _oldctls[name]
_sysctl_set(name, oldval)
if subnets or dnsport:
sysctl_set('net.inet.ip.fw.enable', 1)
changed = sysctl_set('net.inet.ip.scopedroute', 0, permanent=True)
if changed:
log("\n"
" WARNING: ONE-TIME NETWORK DISRUPTION:\n"
" =====================================\n"
"sshuttle has changed a MacOS kernel setting to work around\n"
"a bug in MacOS 10.6. This will cause your network to drop\n"
"within 5-10 minutes unless you restart your network\n"
"interface (change wireless networks or unplug/plug the\n"
"ethernet port) NOW, then restart sshuttle. The fix is\n"
"permanent; you only have to do this once.\n\n")
sys.exit(1)
ipfw('add', sport, 'check-state', 'ip',
'from', 'any', 'to', 'any')
if subnets:
# create new subnet entries
for swidth,sexclude,snet in sorted(subnets, reverse=True):
if sexclude:
ipfw('add', sport, 'skipto', xsport,
'log', 'tcp',
'from', 'any', 'to', '%s/%s' % (snet,swidth))
else:
ipfw('add', sport, 'fwd', '127.0.0.1,%d' % port,
'log', 'tcp',
'from', 'any', 'to', '%s/%s' % (snet,swidth),
'not', 'ipttl', '42', 'keep-state', 'setup')
# This part is much crazier than it is on Linux, because MacOS (at least
# 10.6, and probably other versions, and maybe FreeBSD too) doesn't
# correctly fixup the dstip/dstport for UDP packets when it puts them
# through a 'fwd' rule. It also doesn't fixup the srcip/srcport in the
# response packet. In Linux iptables, all that happens magically for us,
# so we just redirect the packets and relax.
#
# On MacOS, we have to fix the ports ourselves. For that, we use a
# 'divert' socket, which receives raw packets and lets us mangle them.
#
# Here's how it works. Let's say the local DNS server is 1.1.1.1:53,
# and the remote DNS server is 2.2.2.2:53, and the local transproxy port
# is 10.0.0.1:12300, and a client machine is making a request from
# 10.0.0.5:9999. We see a packet like this:
# 10.0.0.5:9999 -> 1.1.1.1:53
# Since the destip:port matches one of our local nameservers, it will
# match a 'fwd' rule, thus grabbing it on the local machine. However,
# the local kernel will then see a packet addressed to *:53 and
# not know what to do with it; there's nobody listening on port 53. Thus,
# we divert it, rewriting it into this:
# 10.0.0.5:9999 -> 10.0.0.1:12300
# This gets proxied out to the server, which sends it to 2.2.2.2:53,
# and the answer comes back, and the proxy sends it back out like this:
# 10.0.0.1:12300 -> 10.0.0.5:9999
# But that's wrong! The original machine expected an answer from
# 1.1.1.1:53, so we have to divert the *answer* and rewrite it:
# 1.1.1.1:53 -> 10.0.0.5:9999
#
# See? Easy stuff.
if dnsport:
divertsock = socket.socket(socket.AF_INET, socket.SOCK_RAW,
IPPROTO_DIVERT)
divertsock.bind(('0.0.0.0', port)) # IP field is ignored
nslist = resolvconf_nameservers()
for ip in nslist:
# relabel and then catch outgoing DNS requests
ipfw('add', sport, 'divert', sport,
'log', 'udp',
'from', 'any', 'to', '%s/32' % ip, '53',
'not', 'ipttl', '42')
# relabel DNS responses
ipfw('add', sport, 'divert', sport,
'log', 'udp',
'from', 'any', str(dnsport), 'to', 'any',
'not', 'ipttl', '42')
def do_wait():
while 1:
r,w,x = select.select([sys.stdin, divertsock], [], [])
if divertsock in r:
_handle_diversion(divertsock, dnsport)
if sys.stdin in r:
return
else:
do_wait = None
return do_wait
def program_exists(name):
paths = (os.getenv('PATH') or os.defpath).split(os.pathsep)
for p in paths:
fn = '%s/%s' % (p, name)
if os.path.exists(fn):
return not os.path.isdir(fn) and os.access(fn, os.X_OK)
hostmap = {}
def rewrite_etc_hosts(port):
HOSTSFILE='/etc/hosts'
BAKFILE='%s.sbak' % HOSTSFILE
APPEND='# sshuttle-firewall-%d AUTOCREATED' % port
old_content = ''
st = None
try:
old_content = open(HOSTSFILE).read()
st = os.stat(HOSTSFILE)
except IOError, e:
if e.errno == errno.ENOENT:
pass
else:
raise
if old_content.strip() and not os.path.exists(BAKFILE):
os.link(HOSTSFILE, BAKFILE)
tmpname = "%s.%d.tmp" % (HOSTSFILE, port)
f = open(tmpname, 'w')
for line in old_content.rstrip().split('\n'):
if line.find(APPEND) >= 0:
continue
f.write('%s\n' % line)
for (name,ip) in sorted(hostmap.items()):
f.write('%-30s %s\n' % ('%s %s' % (ip,name), APPEND))
f.close()
if st:
os.chown(tmpname, st.st_uid, st.st_gid)
os.chmod(tmpname, st.st_mode)
else:
os.chown(tmpname, 0, 0)
os.chmod(tmpname, 0644)
os.rename(tmpname, HOSTSFILE)
def restore_etc_hosts(port):
global hostmap
hostmap = {}
rewrite_etc_hosts(port)
# This is some voodoo for setting up the kernel's transparent
# proxying stuff. If subnets is empty, we just delete our sshuttle rules;
# otherwise we delete it, then make them from scratch.
#
# This code is supposed to clean up after itself by deleting its rules on
# exit. In case that fails, it's not the end of the world; future runs will
# supercede it in the transproxy list, at least, so the leftover rules
# are hopefully harmless.
def main(port, dnsport, syslog):
assert(port > 0)
assert(port <= 65535)
assert(dnsport >= 0)
assert(dnsport <= 65535)
if os.getuid() != 0:
raise Fatal('you must be root (or enable su/sudo) to set the firewall')
if program_exists('ipfw'):
do_it = do_ipfw
elif program_exists('iptables'):
do_it = do_iptables
else:
raise Fatal("can't find either ipfw or iptables; check your PATH")
# because of limitations of the 'su' command, the *real* stdin/stdout
# are both attached to stdout initially. Clone stdout into stdin so we
# can read from it.
os.dup2(1, 0)
if syslog:
ssyslog.start_syslog()
ssyslog.stderr_to_syslog()
debug1('firewall manager ready.\n')
sys.stdout.write('READY\n')
sys.stdout.flush()
# ctrl-c shouldn't be passed along to me. When the main sshuttle dies,
# I'll die automatically.
os.setsid()
# we wait until we get some input before creating the rules. That way,
# sshuttle can launch us as early as possible (and get sudo password
# authentication as early in the startup process as possible).
line = sys.stdin.readline(128)
if not line:
return # parent died; nothing to do
subnets = []
if line != 'ROUTES\n':
raise Fatal('firewall: expected ROUTES but got %r' % line)
while 1:
line = sys.stdin.readline(128)
if not line:
raise Fatal('firewall: expected route but got %r' % line)
elif line == 'GO\n':
break
try:
(width,exclude,ip) = line.strip().split(',', 2)
except:
raise Fatal('firewall: expected route or GO but got %r' % line)
subnets.append((int(width), bool(int(exclude)), ip))
try:
if line:
debug1('firewall manager: starting transproxy.\n')
do_wait = do_it(port, dnsport, subnets)
sys.stdout.write('STARTED\n')
try:
sys.stdout.flush()
except IOError:
# the parent process died for some reason; he's surely been loud
# enough, so no reason to report another error
return
# Now we wait until EOF or any other kind of exception. We need
# to stay running so that we don't need a *second* password
# authentication at shutdown time - that cleanup is important!
while 1:
if do_wait: do_wait()
line = sys.stdin.readline(128)
if line.startswith('HOST '):
(name,ip) = line[5:].strip().split(',', 1)
hostmap[name] = ip
rewrite_etc_hosts(port)
elif line:
raise Fatal('expected EOF, got %r' % line)
else:
break
finally:
try:
debug1('firewall manager: undoing changes.\n')
except:
pass
do_it(port, 0, [])
restore_etc_hosts(port)