mirror of
https://github.com/3proxy/3proxy.git
synced 2026-11-22 23:59:41 +08:00
Transparent moved to main code (TransparentPlugin removed), BSD pf support added
This commit is contained in:
1 parent
85b753ce96
commit
88b3225bdf
19 files changed
+972
-191
No files matched your search
@@ -37,6 +37,7 @@
|
||||
<li><a href="#SSLPLUGIN">How to set up TLS/SSL (https proxy, mTLS)</a></li>
|
||||
<li><a href="#CERTIFICATES">How to create CA and certificates for SSL</a></li>
|
||||
<li><a href="#PCRE">How to use PCRE filtering (regular expressions)</a></li>
|
||||
<li><a href="#TRANSPARENT">How to proxy transparently</a></li>
|
||||
<li><A HREF="#AUTH">How to limit service access</a>
|
||||
<li><A HREF="#USERS">How to create a user list</a>
|
||||
<li><A HREF="#ACL">How to limit user access to resources</a>
|
||||
@@ -981,6 +982,189 @@ openssl pkcs12 -export -out client.p12 -passout pass: \
|
||||
openssl verify -x509_strict -CAfile ca.crt server.crt
|
||||
openssl verify -x509_strict -CAfile ca.crt client.crt
|
||||
</pre>
|
||||
|
||||
<li><a name="TRANSPARENT"><i>How to proxy transparently</i></a>
|
||||
<p>
|
||||
A transparent proxy serves clients that were never configured to use one. A
|
||||
packet filter redirects their connections to 3proxy, and the
|
||||
<b>transparent</b> command tells the service to take the destination from the
|
||||
filter instead of from the request. Every other feature applies as usual:
|
||||
access rules, parent proxies, limits and logging all see the real destination.
|
||||
It works on Linux and on the BSDs. Since 1.0.1 it is part of the binary; before
|
||||
that it was the separate TransparentPlugin, and the <b>plugin</b> line it
|
||||
needed is no longer required.
|
||||
</p>
|
||||
<p>
|
||||
The command supplies both the address and the port the client was trying to
|
||||
reach, so a service can serve whatever was redirected to it rather than one
|
||||
port with one target.
|
||||
</p>
|
||||
<p>
|
||||
Without it a service has to get a destination from somewhere else: an HTTP
|
||||
proxy falls back to the <b>Host</b> header, and a port mapper uses the address
|
||||
it was configured with. Taking the destination from the filter is what makes
|
||||
the other protocols work, and what makes the address authoritative rather than
|
||||
something the client claimed.
|
||||
</p>
|
||||
<p>
|
||||
<b>tlspr</b> is the clearest case. Nothing reaches it at all unless traffic is
|
||||
redirected to it, or the clients resolve the names to it through DNS. With a
|
||||
redirection it gets the address as well as the name from the handshake, and
|
||||
that is what lets access rules be written with host names: the name from the
|
||||
handshake is matched, and the connection still goes to the address the client
|
||||
was going to. Without the address it would have to resolve the name itself,
|
||||
which is a second lookup and a second answer.
|
||||
</p>
|
||||
<p>
|
||||
A configuration for web and TLS traffic:
|
||||
</p><pre>
|
||||
log /var/log/3proxy.log D
|
||||
auth iponly
|
||||
allow *
|
||||
|
||||
# the destination comes from the redirection for the services below
|
||||
transparent
|
||||
|
||||
# ordinary web traffic, redirected here from port 80
|
||||
proxy -p3129 -e192.0.2.10
|
||||
|
||||
# TLS, redirected here from port 443: tlspr would otherwise have only the
|
||||
# name in the handshake, and this gives it the address as well
|
||||
tlspr -p3143 -e192.0.2.10
|
||||
|
||||
notransparent
|
||||
</pre>
|
||||
<p>
|
||||
<b>-e</b> gives those services an address of their own to connect from. That
|
||||
address is what the redirection rules exclude, and excluding it is what stops
|
||||
the proxy's own connections from being redirected back into it. Without an
|
||||
exclusion the connection 3proxy makes to the origin matches the same rule,
|
||||
returns to 3proxy, and the traffic goes round until something gives out.
|
||||
Running 3proxy as its own account and excluding that account works too, and is
|
||||
the better choice when the machine has one address.
|
||||
</p>
|
||||
|
||||
<p><b>Linux, iptables</b>. For traffic the machine forwards for others:
|
||||
</p><pre>
|
||||
iptables -t nat -A PREROUTING -i eth0 -p tcp --dport 80 -j REDIRECT --to-ports 3129
|
||||
iptables -t nat -A PREROUTING -i eth0 -p tcp --dport 443 -j REDIRECT --to-ports 3143
|
||||
</pre>
|
||||
<p>
|
||||
Traffic the machine generates itself passes through OUTPUT instead, where the
|
||||
proxy's own connections have to be excluded:
|
||||
</p><pre>
|
||||
# by the address the services connect from
|
||||
iptables -t nat -A OUTPUT -p tcp --dport 80 ! -s 192.0.2.10 -j REDIRECT --to-ports 3129
|
||||
|
||||
# or by the account 3proxy runs as
|
||||
iptables -t nat -A OUTPUT -p tcp --dport 80 -m owner ! --uid-owner proxy3 \
|
||||
-j REDIRECT --to-ports 3129
|
||||
</pre>
|
||||
|
||||
<p><b>Linux, nftables</b> (Debian 11+, Ubuntu 22.04+, RHEL 8+, Fedora, Arch,
|
||||
where nftables is what iptables is a front end for):
|
||||
</p><pre>
|
||||
table ip proxy3 {
|
||||
chain prerouting {
|
||||
type nat hook prerouting priority dstnat; policy accept;
|
||||
iif "eth0" tcp dport 80 redirect to :3129
|
||||
iif "eth0" tcp dport 443 redirect to :3143
|
||||
}
|
||||
|
||||
chain output {
|
||||
type nat hook output priority dstnat; policy accept;
|
||||
meta skuid != "proxy3" tcp dport 80 redirect to :3129
|
||||
meta skuid != "proxy3" tcp dport 443 redirect to :3143
|
||||
}
|
||||
}
|
||||
</pre>
|
||||
<p>
|
||||
Load it with <b>nft -f</b>, and keep it across reboots in
|
||||
<b>/etc/nftables.conf</b> (Debian, Ubuntu) or
|
||||
<b>/etc/sysconfig/nftables.conf</b> (RHEL, Fedora). A table name cannot begin
|
||||
with a digit, which is why the table above is not called 3proxy.
|
||||
</p>
|
||||
|
||||
<p><b>Linux, firewalld</b> (RHEL, CentOS Stream, Fedora, openSUSE). Redirect an
|
||||
incoming port on a zone:
|
||||
</p><pre>
|
||||
firewall-cmd --permanent --zone=internal --add-forward-port=port=80:proto=tcp:toport=3129
|
||||
firewall-cmd --permanent --zone=internal --add-forward-port=port=443:proto=tcp:toport=3143
|
||||
firewall-cmd --reload
|
||||
</pre>
|
||||
<p>
|
||||
firewalld has no exclusion for the proxy's own traffic in that form, so put
|
||||
that part in a direct rule:
|
||||
</p><pre>
|
||||
firewall-cmd --permanent --direct --add-rule ipv4 nat OUTPUT 0 \
|
||||
-p tcp --dport 80 -m owner ! --uid-owner proxy3 -j REDIRECT --to-ports 3129
|
||||
firewall-cmd --reload
|
||||
</pre>
|
||||
|
||||
<p><b>Linux, ufw</b> (Ubuntu, Debian). ufw has no command for redirection;
|
||||
add the rules to <b>/etc/ufw/before.rules</b>, above the <b>*filter</b> block:
|
||||
</p><pre>
|
||||
*nat
|
||||
:PREROUTING ACCEPT [0:0]
|
||||
-A PREROUTING -i eth0 -p tcp --dport 80 -j REDIRECT --to-ports 3129
|
||||
-A PREROUTING -i eth0 -p tcp --dport 443 -j REDIRECT --to-ports 3143
|
||||
COMMIT
|
||||
</pre>
|
||||
|
||||
<p>
|
||||
On Linux 3proxy asks the kernel where the connection was going. On the BSDs it
|
||||
asks pf, which keeps the original destination in its state table, through
|
||||
<b>/dev/pf</b> - so <b>rdr</b> rules work, and 3proxy has to be able to read
|
||||
that device. Where a redirection leaves the destination on the socket instead,
|
||||
that is used: OpenBSD <b>divert-to</b> and FreeBSD <b>ipfw fwd</b> both do.
|
||||
</p>
|
||||
<p>
|
||||
The mechanism is chosen automatically, and <b>transparent</b> takes an
|
||||
argument for the installations that need to pin it: <b>auto</b> (the default),
|
||||
<b>netfilter</b>, <b>pf</b>, or <b>socket</b> for reading it off the socket. A
|
||||
mode the build has no code for is refused, so a configuration written for
|
||||
another platform fails where it is wrong instead of quietly doing something
|
||||
else.
|
||||
</p>
|
||||
|
||||
<p><b>FreeBSD, NetBSD and OpenBSD, pf</b>. Redirect in <b>/etc/pf.conf</b>,
|
||||
excluding the address the proxy connects from:
|
||||
</p><pre>
|
||||
rdr pass on em0 inet proto tcp from ! 192.0.2.10 to any port 80 -> 127.0.0.1 port 3129
|
||||
rdr pass on em0 inet proto tcp from ! 192.0.2.10 to any port 443 -> 127.0.0.1 port 3143
|
||||
</pre>
|
||||
<p>
|
||||
Load with <b>pfctl -f /etc/pf.conf</b>. 3proxy looks the destination up in pf's
|
||||
state table, so it needs to read <b>/dev/pf</b>: either run it as root, or give
|
||||
its account access to that device.
|
||||
</p>
|
||||
|
||||
<p><b>OpenBSD, divert-to</b> is an alternative which leaves the destination on
|
||||
the socket, and needs no access to <b>/dev/pf</b>:
|
||||
</p><pre>
|
||||
pass in on em0 inet proto tcp to any port 80 divert-to 127.0.0.1 port 3129
|
||||
pass in on em0 inet proto tcp to any port 443 divert-to 127.0.0.1 port 3143
|
||||
</pre>
|
||||
|
||||
<p><b>FreeBSD, ipfw</b>. <b>fwd</b> delivers the connection locally without
|
||||
rewriting it, which also leaves the destination on the socket:
|
||||
</p><pre>
|
||||
ipfw add fwd 127.0.0.1,3129 tcp from any to any 80 in recv em0
|
||||
ipfw add fwd 127.0.0.1,3143 tcp from any to any 443 in recv em0
|
||||
</pre>
|
||||
|
||||
<p><b>macOS</b> has <b>/dev/pf</b> but ships no header for it, so a macOS build
|
||||
has no pf lookup, and macOS has neither <b>divert-to</b> nor ipfw to leave the
|
||||
address on the socket. Transparent proxying is not usable there, even though
|
||||
the commands exist in a macOS build.
|
||||
</p>
|
||||
|
||||
<p>
|
||||
Check the result by asking for a host through a redirected port and reading
|
||||
the log: the request should appear with the address the client asked for,
|
||||
which is what it would look like through a configured proxy.
|
||||
</p>
|
||||
|
||||
<li><a name="PCRE"><i>How to use PCRE filtering (regular expressions)</i></a>
|
||||
<p>
|
||||
Since version 0.9.7, PCRE (Perl Compatible Regular Expressions) filtering is built into
|
||||
|
||||
Reference in new issue
Block a user