In case your FW policy forbids SSH access to the DMZ or internal
network from outside, but you still want to use ssh on machines
which only have one open port, e.g. HTTP, you can use sshttpd
.
sshttpd can multiplex the following protocol pairs:
Be sure you run recent Linux kernel and install nf-conntrack
as well
as libcap
and libcap-devel
if you want to use the capability feature.
On older systems the nf-conntrack module is named nf-conntrack-ipv4.
Inside src
dir:
$ make
There is a new splice
branch inside the git. git checkout splice
before make
, if you want to test this new branch. It implements
zero-copy in terms of the splice(2) system call which has a performance
benefit since it avoids copying the network data between user and kernel
land back and forth (read()/write()), which could also just be spliced kernel-internally
at the "extra cost" of two additional pipe descriptors per connection.
proudly sponsored by:
This paragraph describes the setup where all services run on the same host as sshttpd itself. The muxing happens to the same IP/IP6 address that the outside connects arrive to, so basically just the ports are changing per detected service.
sshttpd is an easy to use OSI-Layer5 switching daemon. It runs
transparently on HTTP(S) port (-L
switch, default 80) and decides
on incoming connections whether this is SSH or HTTP(S) traffic.
If its HTTP(S) traffic, it switches the traffic to the HTTP_PORT
(-H
, default 8080) and if its SSH traffic to SSH_PORT
(-S
, default
22) respectively.
After the build it is time to pick the right netfilter script for setting up the fw rules.
In iptables
directory you will find the old style iptables version and in nft
the new version for nft (but currently only using the compat layer).
You need to edit nf-setup
script (nf6-setup
if using IPv6) to match your network device and $PORTS
(22
and 8080
are just fine for the SSH/HTTP case) and run it to install the proxy rules.
Your sshd has to run on $SSH_PORT
and your webserver on $HTTP_PORT
.
Thats basically it. Go ahead and run sshttpd (as root) and it will layer5-switch
your traffic destinated to TCP port 80:
Take care when running nf-setup on a remote host in an SSH session - it will firewall port 22
# ./nf-setup
Using network device eth0
Setting up port 22 ...
Setting up port 8080 ...
# ./sshttpd -S 22 -L 80 -H 8080 -U nobody -R /var/empty
sshttpd: Using HTTP_PORT=8080 SSH_PORT=22 and local port=80. Going background. Using caps/chroot.
#
If you want to mux SMTP with sshttpd, just give 25
as -L
parameter, 2525
as -H
parameter, and setup your smtp daemon to listen on 2525. Then
edit the nf-setup
script to match these ports. In the Makefile
, change the
SMTP_DOMAIN
and SSH_BANNER
to your needs (SSH_BANNER
must match exactly
yours of the running sshd).
SMTP/SSH muxing was tested with OpenSSH client and Postfix client and server.
When muxing IPv6 connections, the setup is basically the same; just use the nf6-setup
script and invoke sshttpd with -6
.
You can run sshttpd also on your gateway machine and transparently proxy/mux
all of your HTTP(S)/SSH traffic to your internal LAN. To do so, run sshttpd with
-T
and use nf-tproxy
rather than nf-setup
as a template for your FW setup.
Carefully read nf-tproxy
so you don't lock yourself out of the network and all
the network devices and IP addresses match your setup.
With sshttpd you can also mux based on the HTTPS SNI. Just set up your
nf-setup
to contain the SNI ports (there are already samples) and invoke
sshttpd with -N name:port
e.g. sshttpd -S 22 -H 4433 -L 443 -N drops.v2:7350
to hide a sshd on 22 and a drops setup on port 7350 behind port 443, and at the same time serving
your webserver from port 4433 to be visible to outside on port 443.
This works because drops sets the SNI of drops.v2
in outgoing connects.
Multiple -N
switches are allowed so you could mux a lot of services
via SNI. The ports/services must run all on the same machine where the original request
was destinated to. If you just want to mux based on SNI, you can set the SSH port to 0 via -S 0
.
You don't need to patch any of your ssh/web/smtp client or server software. It
works as is. sshttpd runs only on Linux and needs IP_TRANSPARENT
support.
It would work without, but by using IP_TRANSPARENT
it is possible to even
have unmodified syslogs, e.g. the original source IP/port of incoming connections
is passed as-is to the SSH/HTTP/SMTP servers.
Make sure the nf_conntrack
and nf_conntrack_ipv4
or nf_conntrack_ipv6
modules are loaded.
sshttpd is also a tricky anti-SSH0day (if ever:) and anti SSH-scanning/bruteforcing
measurement.
sshttpd has small footprint and was optimized for speed so it also runs
on heavily loaded web servers.
Since version 0.24, sshttpd also supports multiple CPU cores. Unless
-n 1
is used as switch, sshttpd binds one thread per CPU core,
to better exploit the hardware if running on heavily used web servers.
It still runs this fixed number of threads no matter how many 1000s connection
it handles at the same time.
sshttpd runs as nobody
user inside a chroot()
(configurable via -U
and -R
switch)
if compiled with USE_CAPS
. It can also distinguish between SSH and SSL
sessions, you just have to use an LOCAL_PORT (-L)
of 443 or 4433 and change
the HTTP_PORT
in the nf-setup
script to match your webservers HTTPS port.
You cannot mix HTTP/SSH and HTTPS/SSH in one sshttpd instance but you can
run two sshttpd's to reach that goal: one on LOCAL_PORT 80
and one on
LOCAL_PORT 443
.
As per 2017 it seems you have to provide alternative facts for everything, so here are some good writeups from other people for better understanding or in case my description was too brief: