1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
|
<?xml version="1.0"?>
<html>
<body>
<h1 >Firewall and network filtering in libvirt</h1>
<p>There are three pieces of libvirt functionality which do network
filtering of some type.
<br /><br />
At a high level they are:
</p>
<ul>
<li>The virtual network driver
<br /><br />
This provides a isolated bridge device (ie no physical NICs
enslaved). Guest TAP devices are attached to this bridge.
Guests can talk to each other and the host, and optionally the
wider world.
<br /><br />
</li>
<li>The QEMU driver MAC filtering
<br /><br />
This provides a generic filtering of MAC addresses to prevent
the guest spoofing its MAC address. This is mostly obsoleted by
the next item, so won't be discussed further.
<br /><br />
</li>
<li>The network filter driver
<br /><br />
This provides fully configurable, arbitrary network filtering
of traffic on guest NICs. Generic rulesets are defined at the
host level to control traffic in some manner. Rules sets are
then associated with individual NICs of a guest. While not as
expressive as directly using iptables/ebtables, this can still
do nearly everything you would want to on a guest NIC filter.
</li>
</ul>
<h3><a name="name-fw-virtual-network-driver"
id="id-fw-virtual-network-driver">The virtual network driver</a>
</h3>
<p>The typical configuration for guests is to use bridging of the
physical NIC on the host to connect the guest directly to the LAN.
In RHEL6 there is also the possibility of using macvtap/sr-iov
and VEPA connectivity. None of this stuff plays nicely with wireless
NICs, since they will typically silently drop any traffic with a
MAC address that doesn't match that of the physical NIC.
</p>
<p>Thus the virtual network driver in libvirt was invented. This takes
the form of an isolated bridge device (ie one with no physical NICs
enslaved). The TAP devices associated with the guest NICs are attached
to the bridge device. This immediately allows guests on a single host
to talk to each other and to the host OS (modulo host IPtables rules).
</p>
<p>libvirt then uses iptables to control what further connectivity is
available. There are three configurations possible for a virtual
network at time of writing:
</p>
<ul>
<li>isolated: all off-node traffic is completely blocked</li>
<li>nat: outbound traffic to the LAN is allowed, but MASQUERADED</li>
<li>forward: outbound traffic to the LAN is allowed</li>
</ul>
<p>The latter 'forward' case requires the virtual network be on a
separate sub-net from the main LAN, and that the LAN admin has
configured routing for this subnet. In the future we intend to
add support for IP subnetting and/or proxy-arp. This allows for
the virtual network to use the same subnet as the main LAN and
should avoid need for the LAN admin to configure special routing.
</p>
<p>Libvirt will optionally also provide DHCP services to the virtual
network using DNSMASQ. In all cases, we need to allow DNS/DHCP
queries to the host OS. Since we can't predict whether the host
firewall setup is already allowing this, we insert 4 rules into
the head of the INPUT chain
</p>
<pre>
target prot opt in out source destination
ACCEPT udp -- virbr0 * 0.0.0.0/0 0.0.0.0/0 udp dpt:53
ACCEPT tcp -- virbr0 * 0.0.0.0/0 0.0.0.0/0 tcp dpt:53
ACCEPT udp -- virbr0 * 0.0.0.0/0 0.0.0.0/0 udp dpt:67
ACCEPT tcp -- virbr0 * 0.0.0.0/0 0.0.0.0/0 tcp dpt:67</pre>
<p>Note we have restricted our rules to just the bridge associated
with the virtual network, to avoid opening undesirable holes in
the host firewall wrt the LAN/WAN.
</p>
<p>The next rules depend on the type of connectivity allowed, and go
in the main FORWARD chain:
</p>
<ul>
<li>type=isolated
<br /><br />
Allow traffic between guests. Deny inbound. Deny outbound.
<pre>
target prot opt in out source destination
ACCEPT all -- virbr1 virbr1 0.0.0.0/0 0.0.0.0/0
REJECT all -- * virbr1 0.0.0.0/0 0.0.0.0/0 reject-with icmp-port-unreachable
REJECT all -- virbr1 * 0.0.0.0/0 0.0.0.0/0 reject-with icmp-port-unreachable</pre>
</li>
<li>type=nat
<br /><br />
Allow inbound related to an established connection. Allow
outbound, but only from our expected subnet. Allow traffic
between guests. Deny all other inbound. Deny all other outbound.
<pre>
target prot opt in out source destination
ACCEPT all -- * virbr0 0.0.0.0/0 192.168.122.0/24 state RELATED,ESTABLISHED
ACCEPT all -- virbr0 * 192.168.122.0/24 0.0.0.0/0
ACCEPT all -- virbr0 virbr0 0.0.0.0/0 0.0.0.0/0
REJECT all -- * virbr0 0.0.0.0/0 0.0.0.0/0 reject-with icmp-port-unreachable
REJECT all -- virbr0 * 0.0.0.0/0 0.0.0.0/0 reject-with icmp-port-unreachable</pre>
</li>
<li>type=routed
<br /><br />
Allow inbound, but only to our expected subnet. Allow
outbound, but only from our expected subnet. Allow traffic
between guests. Deny all other inbound. Deny all other outbound.
<pre>
target prot opt in out source destination
ACCEPT all -- * virbr2 0.0.0.0/0 192.168.124.0/24
ACCEPT all -- virbr2 * 192.168.124.0/24 0.0.0.0/0
ACCEPT all -- virbr2 virbr2 0.0.0.0/0 0.0.0.0/0
REJECT all -- * virbr2 0.0.0.0/0 0.0.0.0/0 reject-with icmp-port-unreachable
REJECT all -- virbr2 * 0.0.0.0/0 0.0.0.0/0 reject-with icmp-port-unreachable</pre>
</li>
<li>Finally, with type=nat, there is also an entry in the POSTROUTING
chain to apply masquerading:
<pre>
target prot opt in out source destination
MASQUERADE all -- * * 192.168.122.0/24 !192.168.122.0/24</pre>
</li>
</ul>
<h3><a name="name-fw-network-filter-driver"
id="id-fw-network-filter-driver">The network filter driver</a>
</h3>
<p>This driver provides a fully configurable network filtering capability
that leverages ebtables, iptables and ip6tables. This was written by
the libvirt guys at IBM and although its XML schema is defined by libvirt,
the conceptual model is closely aligned with the DMTF CIM schema for
network filtering:
</p>
<p><a href="http://www.dmtf.org/standards/cim/cim_schema_v2230/CIM_Network.pdf">http://www.dmtf.org/standards/cim/cim_schema_v2230/CIM_Network.pdf</a></p>
<p>The filters are managed in libvirt as a top level, standalone object.
This allows the filters to then be referenced by any libvirt object
that requires their functionality, instead tieing them only to use
by guest NICs. In the current implementation, filters can be associated
with individual guest NICs via the libvirt domain XML format. In the
future we might allow filters to be associated with the virtual network
objects. Further we're expecting to define a new 'virtual switch' object
to remove the complexity of configuring bridge/sriov/vepa networking
modes. This make also end up making use of network filters.
</p>
<p>There are a new set of virsh commands for managing network filters:</p>
<ul>
<li>virsh nwfilter-define
<br /><br />
define or update a network filter from an XML file
<br /><br />
</li>
<li>virsh nwfilter-undefine
<br /><br />
undefine a network filter
<br /><br />
</li>
<li>virsh nwfilter-dumpxml
<br /><br />
network filter information in XML
<br /><br />
</li>
<li>virsh nwfilter-list
<br /><br />
list network filters
<br /><br />
</li>
<li>virsh nwfilter-edit
<br /><br />
edit XML configuration for a network filter
</li>
</ul>
<p>There are equivalently named C APIs for each of these commands.</p>
<p>As with all objects libvirt manages, network filters are configured
using an XML format. At a high level the format looks like this:
</p>
<pre>
<filter name='no-spamming' chain='XXXX'>
<uuid>d217f2d7-5a04-0e01-8b98-ec2743436b74</uuid>
<rule ...>
....
</rule>
<filterref filter='XXXX'/>
</filter></pre>
<p>Every filter has a name and UUID which serve as unique identifiers.
A filter can have zero-or-more <code><rule></code> elements which
are used to actually define network controls. Filters can be arranged
into a DAG, so zero-or-more <code><filterref/></code> elements are
also allowed. Cycles in the graph are not allowed.
</p>
<p>The <code><rule></code> element is where all the interesting stuff
happens. It has three attributes, an action, a traffic direction and an
optional priority. eg:
</p>
<pre><rule action='drop' direction='out' priority='500'></pre>
<p>Within the rule there are a wide variety of elements allowed, which
do protocol specific matching. Supported protocols currently include
<code>mac</code>, <code>arp</code>, <code>rarp</code>, <code>ip</code>,
<code>ipv6</code>, <code>tcp/ip</code>, <code>icmp/ip</code>,
<code>igmp/ip</code>, <code>udp/ip</code>, <code>udplite/ip</code>,
<code>esp/ip</code>, <code>ah/ip</code>, <code>sctp/ip</code>,
<code>tcp/ipv6</code>, <code>icmp/ipv6</code>, <code>igmp/ipv6</code>,
<code>udp/ipv6</code>, <code>udplite/ipv6</code>, <code>esp/ipv6</code>,
<code>ah/ipv6</code>, <code>sctp/ipv6</code>. Each protocol defines what
is valid inside the <rule> element. The general pattern though is:
</p>
<pre>
<protocol match='yes|no' attribute1='value1' attribute2='value2'/></pre>
<p>So, eg a TCP protocol, matching ports 0-1023 would be expressed as:</p>
<pre><tcp match='yes' srcportstart='0' srcportend='1023'/></pre>
<p>Attributes can included references to variables defined by the
object using the rule. So the guest XML format allows each NIC
to have a MAC address and IP address defined. These are made
available to filters via the variables <code><b>$IP</b></code> and
<code><b>$MAC</b></code>.
</p>
<p>So to define a filter that prevents IP address spoofing we can
simply match on source IP address <code>!= $IP</code> like this:
</p>
<pre>
<filter name='no-ip-spoofing' chain='ipv4'>
<rule action='drop' direction='out'>
<ip match='no' srcipaddr='<b>$IP</b>' />
</rule>
</filter></pre>
<p>I'm not going to go into details on all the other protocol
matches you can do, because it'll take far too much space.
You can read about the options
<a href="formatnwfilter.html#nwfelemsRulesProto">here</a>.
</p>
<p>Out of the box in RHEL6/Fedora rawhide, libvirt ships with a
set of default useful rules:
</p>
<pre>
# virsh nwfilter-list
UUID Name
----------------------------------------------------------------
15b1ab2b-b1ac-1be2-ed49-2042caba4abb allow-arp
6c51a466-8d14-6d11-46b0-68b1a883d00f allow-dhcp
7517ad6c-bd90-37c8-26c9-4eabcb69848d allow-dhcp-server
3d38b406-7cf0-8335-f5ff-4b9add35f288 allow-incoming-ipv4
5ff06320-9228-2899-3db0-e32554933415 allow-ipv4
db0b1767-d62b-269b-ea96-0cc8b451144e clean-traffic
f88f1932-debf-4aa1-9fbe-f10d3aa4bc95 no-arp-spoofing
772f112d-52e4-700c-0250-e178a3d91a7a no-ip-multicast
7ee20370-8106-765d-f7ff-8a60d5aaf30b no-ip-spoofing
d5d3c490-c2eb-68b1-24fc-3ee362fc8af3 no-mac-broadcast
fb57c546-76dc-a372-513f-e8179011b48a no-mac-spoofing
dba10ea7-446d-76de-346f-335bd99c1d05 no-other-l2-traffic
f5c78134-9da4-0c60-a9f0-fb37bc21ac1f no-other-rarp-traffic
7637e405-4ccf-42ac-5b41-14f8d03d8cf3 qemu-announce-self
9aed52e7-f0f3-343e-fe5c-7dcb27b594e5 qemu-announce-self-rarp</pre>
<p>Most of these are just building blocks. The interesting one here
is 'clean-traffic'. This pulls together all the building blocks
into one filter that you can then associate with a guest NIC.
This stops the most common bad things a guest might try, IP
spoofing, arp spoofing and MAC spoofing. To look at the rules for
any of these just do:
</p>
<pre>virsh nwfilter-dumpxml FILTERNAME|UUID</pre>
<p>They are all stored in <code>/etc/libvirt/nwfilter</code>, but don't
edit the files there directly. Use <code>virsh nwfilter-define</code>
to update them. This ensures the guests have their iptables/ebtables
rules recreated.
</p>
<p>To associate the clean-trafffic filter with a guest, edit the
guest XML config and change the <code><interface></code> element
to include a <code><filterref></code> and also specify the
whitelisted <code><ip address/></code> the guest is allowed to
use:
</p>
<pre>
<interface type='bridge'>
<mac address='52:54:00:56:44:32'/>
<source bridge='br1'/>
<ip address='10.33.8.131'/>
<target dev='vnet0'/>
<model type='virtio'/>
<filterref filter='clean-traffic'/>
</interface></pre>
<p>If no <code><ip address></code> is included, the network filter
driver will activate its 'learning mode'. This uses libpcap to snoop on
network traffic the guest sends and attempts to identify the
first IP address it uses. It then locks traffic to this address.
Obviously this isn't entirely secure, but it does offer some
protection against the guest being trojaned once up and running.
In the future we intend to enhance the learning mode so that it
looks for DHCPOFFERS from a trusted DHCP server and only allows
the offered IP address to be used.
</p>
<p>Now, how is all this implemented...?</p>
<p>The network filter driver uses a combination of ebtables, iptables and
ip6tables, depending on which protocols are referenced in a filter. The
out of the box 'clean-traffic' filter rules only require use of
ebtables. If you want to do matching at tcp/udp/etc protocols (eg to add
a new filter 'no-email-spamming' to block port 25), then iptables will
also be used.
</p>
<p>The driver attempts to keep its rules separate from those that
the host admin might already have configured. So the first thing
it does with ebtables, is to add two hooks in POSTROUTING and
PREROUTING chains, to redirect traffic to custom chains. These
hooks match on the TAP device name of the guest NIC, so they
should not interact badly with any administrator defined rules:
</p>
<pre>
Bridge chain: PREROUTING, entries: 1, policy: ACCEPT
-i vnet0 -j libvirt-I-vnet0
Bridge chain: POSTROUTING, entries: 1, policy: ACCEPT
-o vnet0 -j libvirt-O-vnet0</pre>
<p>To keep things managable and easy to follow, the driver will then
create further sub-chains for each protocol then it needs to match
against:
</p>
<pre>
Bridge chain: libvirt-I-vnet0, entries: 5, policy: ACCEPT
-p IPv4 -j I-vnet0-ipv4
-p ARP -j I-vnet0-arp
-p 0x8035 -j I-vnet0-rarp
-p 0x835 -j ACCEPT
-j DROP
Bridge chain: libvirt-O-vnet0, entries: 4, policy: ACCEPT
-p IPv4 -j O-vnet0-ipv4
-p ARP -j O-vnet0-arp
-p 0x8035 -j O-vnet0-rarp
-j DROP</pre>
<p>Finally, here comes the actual implementation of the filters. This
example shows the 'clean-traffic' filter implementation.
I'm not going to explain what this is doing now. :-)
</p>
<pre>
Bridge chain: I-vnet0-ipv4, entries: 2, policy: ACCEPT
-s ! 52:54:0:56:44:32 -j DROP
-p IPv4 --ip-src ! 10.33.8.131 -j DROP
Bridge chain: O-vnet0-ipv4, entries: 1, policy: ACCEPT
-j ACCEPT
Bridge chain: I-vnet0-arp, entries: 6, policy: ACCEPT
-s ! 52:54:0:56:44:32 -j DROP
-p ARP --arp-mac-src ! 52:54:0:56:44:32 -j DROP
-p ARP --arp-ip-src ! 10.33.8.131 -j DROP
-p ARP --arp-op Request -j ACCEPT
-p ARP --arp-op Reply -j ACCEPT
-j DROP
Bridge chain: O-vnet0-arp, entries: 5, policy: ACCEPT
-p ARP --arp-op Reply --arp-mac-dst ! 52:54:0:56:44:32 -j DROP
-p ARP --arp-ip-dst ! 10.33.8.131 -j DROP
-p ARP --arp-op Request -j ACCEPT
-p ARP --arp-op Reply -j ACCEPT
-j DROP
Bridge chain: I-vnet0-rarp, entries: 2, policy: ACCEPT
-p 0x8035 -s 52:54:0:56:44:32 -d Broadcast --arp-op Request_Reverse --arp-ip-src 0.0.0.0 --arp-ip-dst 0.0.0.0 --arp-mac-src 52:54:0:56:44:32 --arp-mac-dst 52:54:0:56:44:32 -j ACCEPT
-j DROP
Bridge chain: O-vnet0-rarp, entries: 2, policy: ACCEPT
-p 0x8035 -d Broadcast --arp-op Request_Reverse --arp-ip-src 0.0.0.0 --arp-ip-dst 0.0.0.0 --arp-mac-src 52:54:0:56:44:32 --arp-mac-dst 52:54:0:56:44:32 -j ACCEPT
-j DROP</pre>
<p>NB, we would have liked to include the prefix 'libvirt-' in all
of our chain names, but unfortunately the kernel limits names
to a very short maximum length. So only the first two custom
chains can include that prefix. The others just include the
TAP device name + protocol name.
</p>
<p>If I define a new filter 'no-spamming' and then add this to the
'clean-traffic' filter, I can illustrate how iptables usage works:
</p>
<pre>
# cat > /root/spamming.xml <<EOF
<filter name='no-spamming' chain='root'>
<uuid>d217f2d7-5a04-0e01-8b98-ec2743436b74</uuid>
<rule action='drop' direction='out' priority='500'>
<tcp dstportstart='25' dstportend='25'/>
</rule>
</filter>
EOF
# virsh nwfilter-define /root/spamming.xml
# virsh nwfilter-edit clean-traffic</pre>
<p>...add <code><filterref filter='no-spamming'/></code></p>
<p>All active guests immediately have their iptables/ebtables rules
rebuilt.
</p>
<p>The network filter driver deals with iptables in a very similar
way. First it separates out its rules from those the admin may
have defined, by adding a couple of hooks into the INPUT/FORWARD
chains:
</p>
<pre>
Chain INPUT (policy ACCEPT 13M packets, 21G bytes)
target prot opt in out source destination
libvirt-host-in all -- * * 0.0.0.0/0 0.0.0.0/0
Chain FORWARD (policy ACCEPT 5532K packets, 3010M bytes)
target prot opt in out source destination
libvirt-in all -- * * 0.0.0.0/0 0.0.0.0/0
libvirt-out all -- * * 0.0.0.0/0 0.0.0.0/0
libvirt-in-post all -- * * 0.0.0.0/0 0.0.0.0/0</pre>
<p>These custom chains then do matching based on the TAP device
name, so they won't open holes in the admin defined matches for
the LAN/WAN (if any).
</p>
<pre>
Chain libvirt-host-in (1 references)
target prot opt in out source destination
HI-vnet0 all -- * * 0.0.0.0/0 0.0.0.0/0 [goto] PHYSDEV match --physdev-in vnet0
Chain libvirt-in (1 references)
target prot opt in out source destination
FI-vnet0 all -- * * 0.0.0.0/0 0.0.0.0/0 [goto] PHYSDEV match --physdev-in vnet0
Chain libvirt-in-post (1 references)
target prot opt in out source destination
ACCEPT all -- * * 0.0.0.0/0 0.0.0.0/0 PHYSDEV match --physdev-in vnet0
Chain libvirt-out (1 references)
target prot opt in out source destination
FO-vnet0 all -- * * 0.0.0.0/0 0.0.0.0/0 [goto] PHYSDEV match --physdev-out vnet0</pre>
<p>Finally, we can see the interesting bit which is the actual
implementation of my filter to block port 25 access:
</p>
<pre>
Chain FI-vnet0 (1 references)
target prot opt in out source destination
DROP tcp -- * * 0.0.0.0/0 0.0.0.0/0 tcp dpt:25
Chain FO-vnet0 (1 references)
target prot opt in out source destination
DROP tcp -- * * 0.0.0.0/0 0.0.0.0/0 tcp spt:25
Chain HI-vnet0 (1 references)
target prot opt in out source destination
DROP tcp -- * * 0.0.0.0/0 0.0.0.0/0 tcp dpt:25</pre>
<p>One thing in looking at this you may notice is that if there
are many guests all using the same filters, we will be duplicating
the iptables rules over and over for each guest. This is merely a
limitation of the current rules engine implementation. At the libvirt
object modelling level you can clearly see we've designed the model
so filter rules are defined in one place, and indirectly referenced
by guests. Thus it should be possible to change the implementation in
the future so we can share the actual iptables/ebtables rules for
each guest to create a more scalable system. The stuff in current libvirt
is more or less the very first working implementation we've had of this,
so there's not been much optimization work done yet.
</p>
<p>Also notice that at the XML level we don't expose the fact we
are using iptables or ebtables at all. The rule definition is done in
terms of network protocols. Thus if we ever find a need, we could
plug in an alternative implementation that calls out to a different
firewall implementation instead of ebtables/iptables (providing that
implementation was suitably expressive of course)
</p>
<p>Finally, in terms of problems we have in deployment. The biggest
problem is that if the admin does <code>service iptables restart</code>
all our work gets blown away. We've experimented with using lokkit
to record our custom rules in a persistent config file, but that
caused different problem. Admins who were not using lokkit for
their config found that all their own rules got blown away. So
we threw away our lokkit code. Instead we document that if you
run <code>service iptables restart</code>, you need to send SIGHUP to
libvirt to make it recreate its rules.
</p>
<p>More in depth documentation on this is <a href="formatnwfilter.html">here</a>.</p>
</body>
</html>
|
GitWeb