about summary refs log tree commit diff
path: root/doc/dnsfunneld.html
blob: 006a6d68d4deea7644245dffac32d65ea8c10baf (plain) (blame)
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
<html>
  <head>
    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
    <meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />
    <meta http-equiv="Content-Language" content="en" />
    <title>dnsfunnel: the dnsfunnel-daemon program</title>
    <meta name="Description" content="dnsfunnel: the dnsfunnel-daemon program" />
    <meta name="Keywords" content="dnsfunnel daemon /etc/resolv.conf local cache resolver 127.0.0.1" />
    <!-- <link rel="stylesheet" type="text/css" href="//skarnet.org/default.css" /> -->
  </head>
<body>

<p>
<a href="index.html">dnsfunnel</a><br />
<a href="//skarnet.org/software/">Software</a><br />
<a href="//skarnet.org/">skarnet.org</a>
</p>

<h1> The <tt>dnsfunneld</tt> program </h1>

<p>
<tt>dnsfunneld</tt> is a small DNS forwarder daemon. It receives
DNS queries from clients, then forwards them to one or more DNS caches.
It collects the responses and forwards them back to the clients. Depending
on the options it is given, it may perform light processing on the
queries, the responses, or both.
</p>

<h2> Interface </h2>

<pre>
     dnsfunneld [ -v verbosity ] [ -d notif ] [ -o ops ] cachelist
</pre>

<ul>
 <li> dnsfunneld reads the <em>cachelist</em> file, expecting to find
a list of IP (v4 or v6) addresses, one per line. These addresses are the
DNS caches it will forward the queries to. </li>
 <li> dnsfunneld expects to have a bound UDP inet domain socket as
its standard input. It expects to receive packets no more than 512
bytes long, only containing DNS normal queries (QUERY) for the IN
class. </li>
 <li> Depending on <em>ops</em>, dnsfunneld may send additional queries
to the caches listed in <em>cachelist</em>. It handles the answers
internally: the additional queries are invisible to clients. </li>
 <li> dnsfunneld is a long-lived process. </li>
</ul>

<h2> Signals </h2>

<ul>
 <li> SIGHUP: read the <em>cachelist</em> file again, updating its
in-memory cache list. In-flight queries are still handled by the old
list; the new list will only apply for queries arriving after the SIGHUP. </li>
 <li> SIGTERM: enter lame-duck mode, do not accept any more queries. When
all in-flight queries have been answered, exit 0.
</ul>

<h2> Exit codes </h2>

<ul>
 <li> 0: SIGTERM received and all in-flight queries have been answered </li>
 <li> 100: wrong usage </li>
 <li> 111: system call failed </li>
</ul>

<h2> Options </h2>

<ul>
 <li> <tt>-v&nbsp;<em>verbosity</em></tt>&nbsp;: verbosity.
Default is 1. 0 suppresses warning messages. Higher values may give more
informational messages in the future. </li>
 <li> <tt>-d&nbsp;<em>notif</em></tt>&nbsp;: readiness notification. When
dnsfunneld is ready to process queries, write a newline to file descriptor
<em>notif</em>. <em>notif</em> must be 3 or greater. Default is no notification
at all. </li>
 <li> <tt>-o&nbsp;<em>ops</em></tt>&nbsp;: perform various operations on
queries. <em>ops</em> is a decimal integer that is treated as a bitfield.
Default is 0. Operations are listed below. </li>
</ul>

<h2> DNS forwarding behaviour </h2>

<ul>
 <li> When it receives a query, dnsfunneld forwards it to the first DNS cache
in the list it has read from the <em>cachelist</em> file. </li>
 <li> If it receives a response with the TC bit, it resends the query over TCP. </li>
 <li> If it receives a suitable response within a given time frame, it forwards
it to the client. </li>
 <li> On SERVFAIL, or after a timeout of 1 second, it gives up and sends the
query to the next DNS cache in its list. (If the first cache answers after the time
frame, the answer is dropped.)
 <li> If dnsfunneld reaches the end of its cache list, it retries the whole
procedure starting at the beginning of the list, but with a timeout of 3 seconds.
Caches that returned a SERVFAIL are crossed off the list for that query. </li>
 <li> If the second pass fails again, dnsfunneld tries again with a timeout of
11 seconds, then with a timeout of 45 seconds. If all of this fails, it returns
a SERVFAIL to the client. </li>
 <li> A machine should not use a DNS cache that is too far away. In normal operation,
a timeout of 1 second should be more than enough for a cache to answer, if it already
has the answer. If the answer is absent from all caches and it takes them more than
1 second to resolve the query, the answer will be obtained by dnsfunneld in the second
pass. Realistically, the only cases when caches that are not at the top of the list
are used are:
 <ul>
  <li> obscure DNS queries, not likely to be in the caches, and that will take
time to resolve; </li>
  <li> or the first cache has really gone to lunch. </li>
 </ul>
</ul>

<h2> dnsfunneld operations </h2>

<p>
 <em>ops</em> is an integer used as a bitfield. Depending on which bits are set,
various operations are performed on queries or answers, slightly modifying the
behaviour described above.
</p>

<ul>
 <li> bit 0: activate truncation. If a DNS response is more than 510 bytes
long, dnsfunneld will truncate the <em>last</em> resource records in the response,
until it fits into 510 bytes and can be given to the client in a UDP packet.
The structure of a DNS packet makes it so the RRs are listed in order of
decreasing importance, so keeping as many RRs as will fit in 510 bytes
without reordering them is the natural way of truncating a response. </li>
 <li> bit 1: activate workaround for some servers that incorrectly report
NXDOMAIN when they're asked for an AAAA record, and no such record exists
for the domain but an A record exists. When that bit is set in <em>ops</em>,
for every A or AAAA query dnsfunneld receives and forwards, it also sends
an additional AAAA or A query for the same domain. If the main query returns
NXDOMAIN, dnsfunneld waits for the response to the auxiliary query: if this
response is not NXDOMAIN, then dnsfunneld answers NODATA to the client instead
of NXDOMAIN. Be aware that activating this workaround can practically double
the number of queries sent to the DNS caches, and may cause additional delays
before the clients get their answers. </li>
</ul>

<h2> Notes </h2>

<ul>
 <li> The point of dnsfunneld is to work around ill-designed or unreliable
client setups with several motley <tt>nameserver</tt> entries in
<tt>/etc/resolv.conf</tt>. By converting those entries to a cache list
instead (via the <a href="dnsfunnel-translate.html">dnsfunnel-translate</a>
program), running dnsfunneld on 127.0.0.1, and enforcing a policy of one
single <tt>nameserver 127.0.0.1</tt> entry in <tt>/etc/resolv.conf</tt>,
the setup can be made more reliable and more consistent. </li>
 <li> Such a policy can be automated, for instance, by listening to
changes on the <tt>/etc/resolv.conf</tt> file (via inotify or kqueue,
depending on your system) and immediately calling
<a href="dnsfunnel-translate.html">dnsfunnel-translate</a>, sending
a SIGHUP to dnsfunneld, and forcefully overwriting <tt>/etc/resolv.conf</tt>. </li>
 <li> It is easy to send a SIGHUP to dnsfunneld even without knowing its
pid, if it is run under a process supervision system such as
<a href="//skarnet.org/software/s6/">s6</a>. </li>
</ul>

</body>
</html>