[linux] Is there a way for non-root processes to bind to "privileged" ports on Linux?

It's very annoying to have this limitation on my development box, when there won't ever be any users other than me.

I'm aware of the standard workarounds, but none of them do exactly what I want:

1. authbind (The version in Debian testing, 1.0, only supports IPv4)
2. Using the iptables REDIRECT target to redirect a low port to a high port (the "nat" table is not yet implemented for ip6tables, the IPv6 version of iptables)
3. sudo (Running as root is what I'm trying to avoid)
4. SELinux (or similar). (This is just my dev box, I don't want to introduce a lot of extra complexity.)

Is there some simple sysctl variable to allow non-root processes to bind to "privileged" ports (ports less than 1024) on Linux, or am I just out of luck?

EDIT: In some cases, you can use capabilities to do this.

My "standard workaround" uses socat as the user-space redirector:

socat tcp6-listen:80,fork tcp6:8080

Beware that this won't scale, forking is expensive but it's the way socat works.

Use the privbind utility: it allows an unprivileged application to bind to reserved ports.

There is also the 'djb way'. You can use this method to start your process as root running on any port under tcpserver, then it will hand control of the process to the user you specify immediately after the process starts.

#!/bin/sh

UID=$(id -u username)
GID=$(id -g username)
exec tcpserver -u "${UID}" -g "${GID}" -RHl0 0 port /path/to/binary &

For more info, see: http://thedjbway.b0llix.net/daemontools/uidgid.html

Since the OP is just development/testing, less than sleek solutions may be helpful:

setcap can be used on a script's interpreter to grant capabilities to scripts. If setcaps on the global interpreter binary is not acceptable, make a local copy of the binary (any user can) and get root to setcap on this copy. Python2 (at least) works properly with a local copy of the interpreter in your script development tree. No suid is needed so the root user can control to what capabilities users have access.

If you need to track system-wide updates to the interpreter, use a shell script like the following to run your script:

#!/bin/sh
#
#  Watch for updates to the Python2 interpreter

PRG=python_net_raw
PRG_ORIG=/usr/bin/python2.7

cmp $PRG_ORIG $PRG || {
    echo ""
    echo "***** $PRG_ORIG has been updated *****"
    echo "Run the following commands to refresh $PRG:"
    echo ""
    echo "    $ cp $PRG_ORIG $PRG"
    echo "    # setcap cap_net_raw+ep $PRG"
    echo ""
    exit
}

./$PRG $*

Or patch your kernel and remove the check.

(Option of last resort, not recommended).

In net/ipv4/af_inet.c, remove the two lines that read

      if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
              goto out;

and the kernel won't check privileged ports anymore.

Linux supports capabilities to support more fine-grained permissions than just "this application is run as root". One of those capabilities is CAP_NET_BIND_SERVICE which is about binding to a privileged port (<1024).

Unfortunately I don't know how to exploit that to run an application as non-root while still giving it CAP_NET_BIND_SERVICE (probably using setcap, but there's bound to be an existing solution for this).

You can setup a local SSH tunnel, eg if you want port 80 to hit your app bound to 3000:

sudo ssh $USERNAME@localhost -L 80:localhost:3000 -N

This has the advantage of working with script servers, and being very simple.

Modern Linux supports /sbin/sysctl -w net.ipv4.ip_unprivileged_port_start=0.

The standard way is to make them "setuid" so that they start up as root, and then they throw away that root privilege as soon as they've bound to the port but before they start accepting connections to it. You can see good examples of that in the source code for Apache and INN. I'm told that Lighttpd is another good example.

Another example is Postfix, which uses multiple daemons that communicate through pipes, and only one or two of them (which do very little except accept or emit bytes) run as root and the rest run at a lower privilege.

At startup:

iptables -A PREROUTING -t nat -i eth0 -p tcp --dport 80 -j REDIRECT --to-port 8080

Then you can bind to the port you forward to.

For some reason no one mention about lowering sysctl net.ipv4.ip_unprivileged_port_start to the value you need. Example: We need to bind our app to 443 port.

sysctl net.ipv4.ip_unprivileged_port_start=443

Some may say, there is a potential security problem: unprivileged users now may bind to the other privileged ports (444-1024). But you can solve this problem easily with iptables, by blocking other ports:

iptables -I INPUT -p tcp --dport 444:1024 -j DROP
iptables -I INPUT -p udp --dport 444:1024 -j DROP

Comparison with other methods. This method:

• from some point is (IMO) even more secure than setting CAP_NET_BIND_SERVICE/setuid, since an application doesn't setuid at all, even partly (capabilities actually are). For example, to catch a coredump of capability-enabled application you will need to change sysctl fs.suid_dumpable (which leads to another potential security problems) Also, when CAP/suid is set, /proc/PID directory is owned by root, so your non-root user will not have full information/control of running process, for example, user will not be able (in common case) to determine which connections belong to application via /proc/PID/fd/ (netstat -aptn | grep PID).
• has security disadvantage: while your app (or any app that uses ports 443-1024) is down for some reason, another app could take the port. But this problem could also be applied to CAP/suid (in case you set it on interpreter, e.g. java/nodejs) and iptables-redirect. Use systemd-socket method to exclude this problem. Use authbind method to only allow special user binding.
• doesn't require setting CAP/suid every time you deploy new version of application.
• doesn't require application support/modification, like systemd-socket method.
• doesn't require kernel rebuild (if running version supports this sysctl setting)
• doesn't do LD_PRELOAD like authbind/privbind method, this could potentially affect performance, security, behavior (does it? haven't tested). In the rest authbind is really flexible and secure method.
• over-performs iptables REDIRECT/DNAT method, since it doesn't require address translation, connection state tracking, etc. This only noticeable on high-load systems.

Depending on the situation, I would choose between sysctl, CAP, authbind and iptables-redirect. And this is great that we have so many options.

Bind port 8080 to 80 and open port 80:

sudo iptables -t nat -A OUTPUT -o lo -p tcp --dport 80 -j REDIRECT --to-port 8080
sudo iptables -A INPUT -p tcp --dport 80 -j ACCEPT

and then run program on port 8080 as a normal user.

you will then be able to access http://127.0.0.1 on port 80

The standard way is to make them "setuid" so that they start up as root, and then they throw away that root privilege as soon as they've bound to the port but before they start accepting connections to it. You can see good examples of that in the source code for Apache and INN. I'm told that Lighttpd is another good example.

Another example is Postfix, which uses multiple daemons that communicate through pipes, and only one or two of them (which do very little except accept or emit bytes) run as root and the rest run at a lower privilege.

Use the privbind utility: it allows an unprivileged application to bind to reserved ports.

Two other simple possibilities:

There is an old (unfashionable) solution to the "a daemon that binds on a low port and hands control to your daemon". It's called inetd (or xinetd). The cons are:

• your daemon needs to talk on stdin/stdout (if you don't control the daemon -- if you don't have the source -- then this is perhaps a showstopper, although some services may have an inetd-compatibility flag)
• a new daemon process is forked for every connection
• it's one extra link in the chain

Pros:

• available on any old UNIX
• once your sysadmin has set up the config, you're good to go about your development (when you re-build your daemon, might you lose setcap capabilities? And then you'll have to go back to your admin "please sir...")
• daemon doesn't have to worry about that networking stuff, just has to talk on stdin/stdout
• can configure to execute your daemon as a non-root user, as requested

Another alternative: a hacked-up proxy (netcat or even something more robust) from the privileged port to some arbitrary high-numbered port where you can run your target daemon. (Netcat is obviously not a production solution, but "just my dev box", right?). This way you could continue to use a network-capable version of your server, would only need root/sudo to start proxy (at boot), wouldn't be relying on complex/potentially fragile capabilities.

Since the OP is just development/testing, less than sleek solutions may be helpful:

setcap can be used on a script's interpreter to grant capabilities to scripts. If setcaps on the global interpreter binary is not acceptable, make a local copy of the binary (any user can) and get root to setcap on this copy. Python2 (at least) works properly with a local copy of the interpreter in your script development tree. No suid is needed so the root user can control to what capabilities users have access.

If you need to track system-wide updates to the interpreter, use a shell script like the following to run your script:

#!/bin/sh
#
#  Watch for updates to the Python2 interpreter

PRG=python_net_raw
PRG_ORIG=/usr/bin/python2.7

cmp $PRG_ORIG $PRG || {
    echo ""
    echo "***** $PRG_ORIG has been updated *****"
    echo "Run the following commands to refresh $PRG:"
    echo ""
    echo "    $ cp $PRG_ORIG $PRG"
    echo "    # setcap cap_net_raw+ep $PRG"
    echo ""
    exit
}

./$PRG $*

You can do a port redirect. This is what I do for a Silverlight policy server running on a Linux box

iptables -A PREROUTING -t nat -i eth0 -p tcp --dport 943 -j REDIRECT --to-port 1300

Answer at 2015/Sep:

ip6tables now supports IPV6 NAT: http://www.netfilter.org/projects/iptables/files/changes-iptables-1.4.17.txt

You will need kernel 3.7+

Proof:

[09:09:23] root@X:~ ip6tables -t nat -vnL
Chain PREROUTING (policy ACCEPT 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source               destination
    0     0 REDIRECT   tcp      eth0   *       ::/0                 ::/0                 tcp dpt:80 redir ports 8080
    0     0 REDIRECT   tcp      eth0   *       ::/0                 ::/0                 tcp dpt:443 redir ports 1443

Chain INPUT (policy ACCEPT 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source               destination

Chain OUTPUT (policy ACCEPT 6148 packets, 534K bytes)
 pkts bytes target     prot opt in     out     source               destination

Chain POSTROUTING (policy ACCEPT 6148 packets, 534K bytes)
 pkts bytes target     prot opt in     out     source               destination

Two other simple possibilities:

There is an old (unfashionable) solution to the "a daemon that binds on a low port and hands control to your daemon". It's called inetd (or xinetd). The cons are:

• your daemon needs to talk on stdin/stdout (if you don't control the daemon -- if you don't have the source -- then this is perhaps a showstopper, although some services may have an inetd-compatibility flag)
• a new daemon process is forked for every connection
• it's one extra link in the chain

Pros:

• available on any old UNIX
• once your sysadmin has set up the config, you're good to go about your development (when you re-build your daemon, might you lose setcap capabilities? And then you'll have to go back to your admin "please sir...")
• daemon doesn't have to worry about that networking stuff, just has to talk on stdin/stdout
• can configure to execute your daemon as a non-root user, as requested

Another alternative: a hacked-up proxy (netcat or even something more robust) from the privileged port to some arbitrary high-numbered port where you can run your target daemon. (Netcat is obviously not a production solution, but "just my dev box", right?). This way you could continue to use a network-capable version of your server, would only need root/sudo to start proxy (at boot), wouldn't be relying on complex/potentially fragile capabilities.

At startup:

iptables -A PREROUTING -t nat -i eth0 -p tcp --dport 80 -j REDIRECT --to-port 8080

Then you can bind to the port you forward to.

Answer at 2015/Sep:

ip6tables now supports IPV6 NAT: http://www.netfilter.org/projects/iptables/files/changes-iptables-1.4.17.txt

You will need kernel 3.7+

Proof:

[09:09:23] root@X:~ ip6tables -t nat -vnL
Chain PREROUTING (policy ACCEPT 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source               destination
    0     0 REDIRECT   tcp      eth0   *       ::/0                 ::/0                 tcp dpt:80 redir ports 8080
    0     0 REDIRECT   tcp      eth0   *       ::/0                 ::/0                 tcp dpt:443 redir ports 1443

Chain INPUT (policy ACCEPT 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source               destination

Chain OUTPUT (policy ACCEPT 6148 packets, 534K bytes)
 pkts bytes target     prot opt in     out     source               destination

Chain POSTROUTING (policy ACCEPT 6148 packets, 534K bytes)
 pkts bytes target     prot opt in     out     source               destination

Linux supports capabilities to support more fine-grained permissions than just "this application is run as root". One of those capabilities is CAP_NET_BIND_SERVICE which is about binding to a privileged port (<1024).

Unfortunately I don't know how to exploit that to run an application as non-root while still giving it CAP_NET_BIND_SERVICE (probably using setcap, but there's bound to be an existing solution for this).

File capabilities are not ideal, because they can break after a package update.

The ideal solution, IMHO, should be an ability to create a shell with inheritable CAP_NET_BIND_SERVICE set.

Here's a somewhat convoluted way to do this:

sg $DAEMONUSER "capsh --keep=1 --uid=`id -u $DAEMONUSER` \
     --caps='cap_net_bind_service+pei' -- \
     YOUR_COMMAND_GOES_HERE"

capsh utility can be found in libcap2-bin package in Debian/Ubuntu distributions. Here's what goes on:

• sg changes effective group ID to that of the daemon user. This is necessary because capsh leaves GID unchanged and we definitely do not want it.
• Sets bit 'keep capabilities on UID change'.
• Changes UID to $DAEMONUSER
• Drops all caps (at this moment all caps are still present because of --keep=1), except inheritable cap_net_bind_service
• Executes your command ('--' is a separator)

The result is a process with specified user and group, and cap_net_bind_service privileges.

As an example, a line from ejabberd startup script:

sg $EJABBERDUSER "capsh --keep=1 --uid=`id -u $EJABBERDUSER` --caps='cap_net_bind_service+pei' -- $EJABBERD --noshell -detached"

My "standard workaround" uses socat as the user-space redirector:

socat tcp6-listen:80,fork tcp6:8080

Beware that this won't scale, forking is expensive but it's the way socat works.

I tried the iptables PREROUTING REDIRECT method. In older kernels it seems this type of rule wasn't supported for IPv6. But apparently it is now supported in ip6tables v1.4.18 and Linux kernel v3.8.

I also found that PREROUTING REDIRECT doesn't work for connections initiated within the machine. To work for conections from the local machine, add an OUTPUT rule also — see iptables port redirect not working for localhost. E.g. something like:

iptables -t nat -I OUTPUT -o lo -p tcp --dport 80 -j REDIRECT --to-port 8080

I also found that PREROUTING REDIRECT also affects forwarded packets. That is, if the machine is also forwarding packets between interfaces (e.g. if it's acting as a Wi-Fi access point connected to an Ethernet network), then the iptables rule will also catch connected clients' connections to Internet destinations, and redirect them to the machine. That's not what I wanted—I only wanted to redirect connections that were directed to the machine itself. I found I can make it only affect packets addressed to the box, by adding -m addrtype --dst-type LOCAL. E.g. something like:

iptables -A PREROUTING -t nat -p tcp --dport 80 -m addrtype --dst-type LOCAL -j REDIRECT --to-port 8080

One other possibility is to use TCP port forwarding. E.g. using socat:

socat TCP4-LISTEN:www,reuseaddr,fork TCP4:localhost:8080

However one disadvantage with that method is, the application that is listening on port 8080 then doesn't know the source address of incoming connections (e.g. for logging or other identification purposes).

File capabilities are not ideal, because they can break after a package update.

The ideal solution, IMHO, should be an ability to create a shell with inheritable CAP_NET_BIND_SERVICE set.

Here's a somewhat convoluted way to do this:

sg $DAEMONUSER "capsh --keep=1 --uid=`id -u $DAEMONUSER` \
     --caps='cap_net_bind_service+pei' -- \
     YOUR_COMMAND_GOES_HERE"

capsh utility can be found in libcap2-bin package in Debian/Ubuntu distributions. Here's what goes on:

• sg changes effective group ID to that of the daemon user. This is necessary because capsh leaves GID unchanged and we definitely do not want it.
• Sets bit 'keep capabilities on UID change'.
• Changes UID to $DAEMONUSER
• Drops all caps (at this moment all caps are still present because of --keep=1), except inheritable cap_net_bind_service
• Executes your command ('--' is a separator)

The result is a process with specified user and group, and cap_net_bind_service privileges.

As an example, a line from ejabberd startup script:

sg $EJABBERDUSER "capsh --keep=1 --uid=`id -u $EJABBERDUSER` --caps='cap_net_bind_service+pei' -- $EJABBERD --noshell -detached"

Port redirect made the most sense for us, but we ran into an issue where our application would resolve a url locally that also needed to be re-routed; (that means you shindig).

This will also allow you to be redirected when accessing the url on the local machine.

iptables -A PREROUTING -t nat -p tcp --dport 80 -j REDIRECT --to-port 8080
iptables -A OUTPUT -t nat -p tcp --dport 80 -j REDIRECT --to-port 8080

Or patch your kernel and remove the check.

(Option of last resort, not recommended).

In net/ipv4/af_inet.c, remove the two lines that read

      if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
              goto out;

and the kernel won't check privileged ports anymore.

I know this is an old question, but now with recent (>= 4.3) kernels there is finally a good answer to this - ambient capabilities.

The quick answer is to grab a copy of the latest (as-yet-unreleased) version of libcap from git and compile it. Copy the resulting progs/capsh binary somewhere (/usr/local/bin is a good choice). Then, as root, start your program with

/usr/local/bin/capsh --keep=1 --user='your-service-user-name' \
    --inh='cap_net_bind_service' --addamb='cap_net_bind_service' \ 
    -- -c 'your-program'

In order, we are

• Declaring that when we switch users, we want to keep our current capability sets
• Switching user & group to 'your-service-user-name'
• Adding the cap_net_bind_service capability to the inherited & ambient sets
• Forking bash -c 'your-command' (since capsh automatically starts bash with the arguments after --)

There's a lot going on under the hood here.

Firstly, we are running as root, so by default, we get a full set of capabilities. Included in this is the ability to switch uid & gid with the setuid and setgid syscalls. However, ordinarily when a program does this, it loses its set of capabilities - this is so that the old way of dropping root with setuid still works. The --keep=1 flag tells capsh to issue the prctl(PR_SET_KEEPCAPS) syscall, which disables the dropping of capabilities when changing user. The actual changing of users by capsh happens with the --user flag, which runs setuid and setgid.

The next problem we need to solve is how to set capabilities in a way that carries on after we exec our children. The capabilities system has always had an 'inherited' set of capabilities, which is " a set of capabilities preserved across an execve(2)" [capabilities(7)]. Whilst this sounds like it solves our problem (just set the cap_net_bind_service capability to inherited, right?), this actually only applies for privileged processes - and our process is not privileged anymore, because we already changed user (with the --user flag).

The new ambient capability set works around this problem - it is "a set of capabilities that are preserved across an execve(2) of a program that is not privileged." By putting cap_net_bind_service in the ambient set, when capsh exec's our server program, our program will inherit this capability and be able to bind listeners to low ports.

If you're interested to learn more, the capabilities manual page explains this in great detail. Running capsh through strace is also very informative!

Two other simple possibilities:

There is an old (unfashionable) solution to the "a daemon that binds on a low port and hands control to your daemon". It's called inetd (or xinetd). The cons are:

• your daemon needs to talk on stdin/stdout (if you don't control the daemon -- if you don't have the source -- then this is perhaps a showstopper, although some services may have an inetd-compatibility flag)
• a new daemon process is forked for every connection
• it's one extra link in the chain

Pros:

• available on any old UNIX
• once your sysadmin has set up the config, you're good to go about your development (when you re-build your daemon, might you lose setcap capabilities? And then you'll have to go back to your admin "please sir...")
• daemon doesn't have to worry about that networking stuff, just has to talk on stdin/stdout
• can configure to execute your daemon as a non-root user, as requested

Another alternative: a hacked-up proxy (netcat or even something more robust) from the privileged port to some arbitrary high-numbered port where you can run your target daemon. (Netcat is obviously not a production solution, but "just my dev box", right?). This way you could continue to use a network-capable version of your server, would only need root/sudo to start proxy (at boot), wouldn't be relying on complex/potentially fragile capabilities.

With systemd, you just need to slightly modify your service to accept preactivated sockets.

You can later use systemd socket activate.

No capabilities, iptables or other tricks are needed.

This is content of relevant systemd files from this example of simple python http server

File httpd-true.service

[Unit]
Description=Httpd true 

[Service]
ExecStart=/usr/local/bin/httpd-true
User=subsonic

PrivateTmp=yes

File httpd-true.socket

[Unit]
Description=HTTPD true

[Socket]
ListenStream=80

[Install]
WantedBy=default.target

Two other simple possibilities:

There is an old (unfashionable) solution to the "a daemon that binds on a low port and hands control to your daemon". It's called inetd (or xinetd). The cons are:

• your daemon needs to talk on stdin/stdout (if you don't control the daemon -- if you don't have the source -- then this is perhaps a showstopper, although some services may have an inetd-compatibility flag)
• a new daemon process is forked for every connection
• it's one extra link in the chain

Pros:

• available on any old UNIX
• once your sysadmin has set up the config, you're good to go about your development (when you re-build your daemon, might you lose setcap capabilities? And then you'll have to go back to your admin "please sir...")
• daemon doesn't have to worry about that networking stuff, just has to talk on stdin/stdout
• can configure to execute your daemon as a non-root user, as requested

Another alternative: a hacked-up proxy (netcat or even something more robust) from the privileged port to some arbitrary high-numbered port where you can run your target daemon. (Netcat is obviously not a production solution, but "just my dev box", right?). This way you could continue to use a network-capable version of your server, would only need root/sudo to start proxy (at boot), wouldn't be relying on complex/potentially fragile capabilities.

TLDR: For "the answer" (as I see it), jump down to the >>TLDR<< part in this answer.

OK, I've figured it out (for real this time), the answer to this question, and this answer of mine is also a way of apologizing for promoting another answer (both here and on twitter) that I thought was "the best", but after trying it, discovered that I was mistaken about that. Learn from my mistake kids: don't promote something until you've actually tried it yourself!

Again, I reviewed all the answers here. I've tried some of them (and chose not to try others because I simply didn't like the solutions). I thought that the solution was to use systemd with its Capabilities= and CapabilitiesBindingSet= settings. After wrestling with this for some time, I discovered that this is not the solution because:

Capabilities are intended to restrict root processes!

As the OP wisely stated, it is always best to avoid that (for all your daemons if possible!).

You cannot use the Capabilities related options with User= and Group= in systemd unit files, because capabilities are ALWAYS reset when execev (or whatever the function is) is called. In other words, when systemd forks and drops its perms, the capabilities are reset. There is no way around this, and all that binding logic in the kernel is basic around uid=0, not capabilities. This means that it is unlikely that Capabilities will ever be the right answer to this question (at least any time soon). Incidentally, setcap, as others have mentioned, is not a solution. It didn't work for me, it doesn't work nicely with scripts, and those are reset anyways whenever the file changes.

In my meager defense, I did state (in the comment I've now deleted), that James' iptables suggestion (which the OP also mentions), was the "2nd best solution". :-P

>>TLDR<<

The solution is to combine systemd with on-the-fly iptables commands, like this (taken from DNSChain):

[Unit]
Description=dnschain
After=network.target
Wants=namecoin.service

[Service]
ExecStart=/usr/local/bin/dnschain
Environment=DNSCHAIN_SYSD_VER=0.0.1
PermissionsStartOnly=true
ExecStartPre=/sbin/sysctl -w net.ipv4.ip_forward=1
ExecStartPre=-/sbin/iptables -D INPUT -p udp --dport 5333 -j ACCEPT
ExecStartPre=-/sbin/iptables -t nat -D PREROUTING -p udp --dport 53 -j REDIRECT --to-ports 5333
ExecStartPre=/sbin/iptables -A INPUT -p udp --dport 5333 -j ACCEPT
ExecStartPre=/sbin/iptables -t nat -A PREROUTING -p udp --dport 53 -j REDIRECT --to-ports 5333
ExecStopPost=/sbin/iptables -D INPUT -p udp --dport 5333 -j ACCEPT
ExecStopPost=/sbin/iptables -t nat -D PREROUTING -p udp --dport 53 -j REDIRECT --to-ports 5333
User=dns
Group=dns
Restart=always
RestartSec=5
WorkingDirectory=/home/dns
PrivateTmp=true
NoNewPrivileges=true
ReadOnlyDirectories=/etc

# Unfortunately, capabilities are basically worthless because they're designed to restrict root daemons. Instead, we use iptables to listen on privileged ports.
# Capabilities=cap_net_bind_service+pei
# SecureBits=keep-caps

[Install]
WantedBy=multi-user.target

Here we accomplish the following:

• The daemon listens on 5333, but connections are successfully accepted on 53 thanks to iptables
• We can include the commands in the unit file itself, and thus we save people headaches. systemd cleans up the firewall rules for us, making sure to remove them when the daemon isn't running.
• We never run as root, and we make privilege escalation impossible (at least systemd claims to), supposedly even if the daemon is compromised and sets uid=0.

iptables is still, unfortunately, quite an ugly and difficult-to-use utility. If the daemon is listening on eth0:0 instead of eth0, for example, the commands are slightly different.

Port redirect made the most sense for us, but we ran into an issue where our application would resolve a url locally that also needed to be re-routed; (that means you shindig).

This will also allow you to be redirected when accessing the url on the local machine.

iptables -A PREROUTING -t nat -p tcp --dport 80 -j REDIRECT --to-port 8080
iptables -A OUTPUT -t nat -p tcp --dport 80 -j REDIRECT --to-port 8080

With systemd, you just need to slightly modify your service to accept preactivated sockets.

You can later use systemd socket activate.

No capabilities, iptables or other tricks are needed.

This is content of relevant systemd files from this example of simple python http server

File httpd-true.service

[Unit]
Description=Httpd true 

[Service]
ExecStart=/usr/local/bin/httpd-true
User=subsonic

PrivateTmp=yes

File httpd-true.socket

[Unit]
Description=HTTPD true

[Socket]
ListenStream=80

[Install]
WantedBy=default.target

There is also the 'djb way'. You can use this method to start your process as root running on any port under tcpserver, then it will hand control of the process to the user you specify immediately after the process starts.

#!/bin/sh

UID=$(id -u username)
GID=$(id -g username)
exec tcpserver -u "${UID}" -g "${GID}" -RHl0 0 port /path/to/binary &

For more info, see: http://thedjbway.b0llix.net/daemontools/uidgid.html

Linux supports capabilities to support more fine-grained permissions than just "this application is run as root". One of those capabilities is CAP_NET_BIND_SERVICE which is about binding to a privileged port (<1024).

Unfortunately I don't know how to exploit that to run an application as non-root while still giving it CAP_NET_BIND_SERVICE (probably using setcap, but there's bound to be an existing solution for this).

I know this is an old question, but now with recent (>= 4.3) kernels there is finally a good answer to this - ambient capabilities.

The quick answer is to grab a copy of the latest (as-yet-unreleased) version of libcap from git and compile it. Copy the resulting progs/capsh binary somewhere (/usr/local/bin is a good choice). Then, as root, start your program with

/usr/local/bin/capsh --keep=1 --user='your-service-user-name' \
    --inh='cap_net_bind_service' --addamb='cap_net_bind_service' \ 
    -- -c 'your-program'

In order, we are

• Declaring that when we switch users, we want to keep our current capability sets
• Switching user & group to 'your-service-user-name'
• Adding the cap_net_bind_service capability to the inherited & ambient sets
• Forking bash -c 'your-command' (since capsh automatically starts bash with the arguments after --)

There's a lot going on under the hood here.

Firstly, we are running as root, so by default, we get a full set of capabilities. Included in this is the ability to switch uid & gid with the setuid and setgid syscalls. However, ordinarily when a program does this, it loses its set of capabilities - this is so that the old way of dropping root with setuid still works. The --keep=1 flag tells capsh to issue the prctl(PR_SET_KEEPCAPS) syscall, which disables the dropping of capabilities when changing user. The actual changing of users by capsh happens with the --user flag, which runs setuid and setgid.

The next problem we need to solve is how to set capabilities in a way that carries on after we exec our children. The capabilities system has always had an 'inherited' set of capabilities, which is " a set of capabilities preserved across an execve(2)" [capabilities(7)]. Whilst this sounds like it solves our problem (just set the cap_net_bind_service capability to inherited, right?), this actually only applies for privileged processes - and our process is not privileged anymore, because we already changed user (with the --user flag).

The new ambient capability set works around this problem - it is "a set of capabilities that are preserved across an execve(2) of a program that is not privileged." By putting cap_net_bind_service in the ambient set, when capsh exec's our server program, our program will inherit this capability and be able to bind listeners to low ports.

If you're interested to learn more, the capabilities manual page explains this in great detail. Running capsh through strace is also very informative!

I tried the iptables PREROUTING REDIRECT method. In older kernels it seems this type of rule wasn't supported for IPv6. But apparently it is now supported in ip6tables v1.4.18 and Linux kernel v3.8.

I also found that PREROUTING REDIRECT doesn't work for connections initiated within the machine. To work for conections from the local machine, add an OUTPUT rule also — see iptables port redirect not working for localhost. E.g. something like:

iptables -t nat -I OUTPUT -o lo -p tcp --dport 80 -j REDIRECT --to-port 8080

I also found that PREROUTING REDIRECT also affects forwarded packets. That is, if the machine is also forwarding packets between interfaces (e.g. if it's acting as a Wi-Fi access point connected to an Ethernet network), then the iptables rule will also catch connected clients' connections to Internet destinations, and redirect them to the machine. That's not what I wanted—I only wanted to redirect connections that were directed to the machine itself. I found I can make it only affect packets addressed to the box, by adding -m addrtype --dst-type LOCAL. E.g. something like:

iptables -A PREROUTING -t nat -p tcp --dport 80 -m addrtype --dst-type LOCAL -j REDIRECT --to-port 8080

One other possibility is to use TCP port forwarding. E.g. using socat:

socat TCP4-LISTEN:www,reuseaddr,fork TCP4:localhost:8080

However one disadvantage with that method is, the application that is listening on port 8080 then doesn't know the source address of incoming connections (e.g. for logging or other identification purposes).

Linux supports capabilities to support more fine-grained permissions than just "this application is run as root". One of those capabilities is CAP_NET_BIND_SERVICE which is about binding to a privileged port (<1024).

Unfortunately I don't know how to exploit that to run an application as non-root while still giving it CAP_NET_BIND_SERVICE (probably using setcap, but there's bound to be an existing solution for this).

Update 2017:

Use authbind

• CAP_NET_BIND_SERVICE grants trust to the binary but provides no control over per-port access.
  

• Authbind grants trust to the user/group and provides control over per-port access, and supports both IPv4 and IPv6 (IPv6 support has been added as of late).

  1. Install: apt-get install authbind

  2. Configure access to relevant ports, e.g. 80 and 443 for all users and groups:

     sudo touch /etc/authbind/byport/80
     sudo touch /etc/authbind/byport/443
     sudo chmod 777 /etc/authbind/byport/80
     sudo chmod 777 /etc/authbind/byport/443
     

  3. Execute your command via authbind
     (optionally specifying --deep or other arguments, see the man page):

     authbind --deep /path/to/binary command line args
     

     e.g.

     authbind --deep java -jar SomeServer.jar
     

As a follow-up to Joshua's fabulous (=not recommended unless you know what you do) recommendation to hack the kernel:

I've first posted it here.

Simple. With a normal or old kernel, you don't.
As pointed out by others, iptables can forward a port.
As also pointed out by others, CAP_NET_BIND_SERVICE can also do the job.
Of course CAP_NET_BIND_SERVICE will fail if you launch your program from a script, unless you set the cap on the shell interpreter, which is pointless, you could just as well run your service as root...
e.g. for Java, you have to apply it to the JAVA JVM

sudo /sbin/setcap 'cap_net_bind_service=ep' /usr/lib/jvm/java-8-openjdk/jre/bin/java

Obviously, that then means any Java program can bind system ports.
Dito for mono/.NET.

I'm also pretty sure xinetd isn't the best of ideas.
But since both methods are hacks, why not just lift the limit by lifting the restriction ?
Nobody said you have to run a normal kernel, so you can just run your own.

You just download the source for the latest kernel (or the same you currently have). Afterwards, you go to:

/usr/src/linux-<version_number>/include/net/sock.h:

There you look for this line

/* Sockets 0-1023 can't be bound to unless you are superuser */
#define PROT_SOCK       1024

and change it to

#define PROT_SOCK 0

if you don't want to have an insecure ssh situation, you alter it to this: #define PROT_SOCK 24

Generally, I'd use the lowest setting that you need, e.g 79 for http, or 24 when using SMTP on port 25.

That's already all.
Compile the kernel, and install it.
Reboot.
Finished - that stupid limit is GONE, and that also works for scripts.

Here's how you compile a kernel:

https://help.ubuntu.com/community/Kernel/Compile

# You can get the kernel-source via package linux-source, no manual download required
apt-get install linux-source fakeroot

mkdir ~/src
cd ~/src
tar xjvf /usr/src/linux-source-<version>.tar.bz2
cd linux-source-<version>

# Apply the changes to PROT_SOCK define in /include/net/sock.h

# Copy the kernel config file you are currently using
cp -vi /boot/config-`uname -r` .config

# Install ncurses libary, if you want to run menuconfig
apt-get install libncurses5 libncurses5-dev

# Run menuconfig (optional)
make menuconfig

# Define the number of threads you wanna use when compiling (should be <number CPU cores> - 1), e.g. for quad-core
export CONCURRENCY_LEVEL=3
# Now compile the custom kernel
fakeroot make-kpkg --initrd --append-to-version=custom kernel-image kernel-headers

# And wait a long long time

cd ..

In a nutshell, use iptables if you want to stay secure, compile the kernel if you want to be sure this restriction never bothers you again.

The standard way is to make them "setuid" so that they start up as root, and then they throw away that root privilege as soon as they've bound to the port but before they start accepting connections to it. You can see good examples of that in the source code for Apache and INN. I'm told that Lighttpd is another good example.

Another example is Postfix, which uses multiple daemons that communicate through pipes, and only one or two of them (which do very little except accept or emit bytes) run as root and the rest run at a lower privilege.

systemd is a sysvinit replacement which has an option to launch a daemon with specific capabilities. Options Capabilities=, CapabilityBoundingSet= in systemd.exec(5) manpage.

Or patch your kernel and remove the check.

(Option of last resort, not recommended).

In net/ipv4/af_inet.c, remove the two lines that read

      if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
              goto out;

and the kernel won't check privileged ports anymore.

You can do a port redirect. This is what I do for a Silverlight policy server running on a Linux box

iptables -A PREROUTING -t nat -i eth0 -p tcp --dport 943 -j REDIRECT --to-port 1300

Update 2017:

Use authbind

• CAP_NET_BIND_SERVICE grants trust to the binary but provides no control over per-port access.
  

• Authbind grants trust to the user/group and provides control over per-port access, and supports both IPv4 and IPv6 (IPv6 support has been added as of late).

  1. Install: apt-get install authbind

  2. Configure access to relevant ports, e.g. 80 and 443 for all users and groups:

     sudo touch /etc/authbind/byport/80
     sudo touch /etc/authbind/byport/443
     sudo chmod 777 /etc/authbind/byport/80
     sudo chmod 777 /etc/authbind/byport/443
     

  3. Execute your command via authbind
     (optionally specifying --deep or other arguments, see the man page):

     authbind --deep /path/to/binary command line args
     

     e.g.

     authbind --deep java -jar SomeServer.jar
     

As a follow-up to Joshua's fabulous (=not recommended unless you know what you do) recommendation to hack the kernel:

I've first posted it here.

Simple. With a normal or old kernel, you don't.
As pointed out by others, iptables can forward a port.
As also pointed out by others, CAP_NET_BIND_SERVICE can also do the job.
Of course CAP_NET_BIND_SERVICE will fail if you launch your program from a script, unless you set the cap on the shell interpreter, which is pointless, you could just as well run your service as root...
e.g. for Java, you have to apply it to the JAVA JVM

sudo /sbin/setcap 'cap_net_bind_service=ep' /usr/lib/jvm/java-8-openjdk/jre/bin/java

Obviously, that then means any Java program can bind system ports.
Dito for mono/.NET.

I'm also pretty sure xinetd isn't the best of ideas.
But since both methods are hacks, why not just lift the limit by lifting the restriction ?
Nobody said you have to run a normal kernel, so you can just run your own.

You just download the source for the latest kernel (or the same you currently have). Afterwards, you go to:

/usr/src/linux-<version_number>/include/net/sock.h:

There you look for this line

/* Sockets 0-1023 can't be bound to unless you are superuser */
#define PROT_SOCK       1024

and change it to

#define PROT_SOCK 0

if you don't want to have an insecure ssh situation, you alter it to this: #define PROT_SOCK 24

Generally, I'd use the lowest setting that you need, e.g 79 for http, or 24 when using SMTP on port 25.

That's already all.
Compile the kernel, and install it.
Reboot.
Finished - that stupid limit is GONE, and that also works for scripts.

Here's how you compile a kernel:

https://help.ubuntu.com/community/Kernel/Compile

# You can get the kernel-source via package linux-source, no manual download required
apt-get install linux-source fakeroot

mkdir ~/src
cd ~/src
tar xjvf /usr/src/linux-source-<version>.tar.bz2
cd linux-source-<version>

# Apply the changes to PROT_SOCK define in /include/net/sock.h

# Copy the kernel config file you are currently using
cp -vi /boot/config-`uname -r` .config

# Install ncurses libary, if you want to run menuconfig
apt-get install libncurses5 libncurses5-dev

# Run menuconfig (optional)
make menuconfig

# Define the number of threads you wanna use when compiling (should be <number CPU cores> - 1), e.g. for quad-core
export CONCURRENCY_LEVEL=3
# Now compile the custom kernel
fakeroot make-kpkg --initrd --append-to-version=custom kernel-image kernel-headers

# And wait a long long time

cd ..

In a nutshell, use iptables if you want to stay secure, compile the kernel if you want to be sure this restriction never bothers you again.

Or patch your kernel and remove the check.

(Option of last resort, not recommended).

In net/ipv4/af_inet.c, remove the two lines that read

      if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
              goto out;

and the kernel won't check privileged ports anymore.

For some reason no one mention about lowering sysctl net.ipv4.ip_unprivileged_port_start to the value you need. Example: We need to bind our app to 443 port.

sysctl net.ipv4.ip_unprivileged_port_start=443

Some may say, there is a potential security problem: unprivileged users now may bind to the other privileged ports (444-1024). But you can solve this problem easily with iptables, by blocking other ports:

iptables -I INPUT -p tcp --dport 444:1024 -j DROP
iptables -I INPUT -p udp --dport 444:1024 -j DROP

Comparison with other methods. This method:

• from some point is (IMO) even more secure than setting CAP_NET_BIND_SERVICE/setuid, since an application doesn't setuid at all, even partly (capabilities actually are). For example, to catch a coredump of capability-enabled application you will need to change sysctl fs.suid_dumpable (which leads to another potential security problems) Also, when CAP/suid is set, /proc/PID directory is owned by root, so your non-root user will not have full information/control of running process, for example, user will not be able (in common case) to determine which connections belong to application via /proc/PID/fd/ (netstat -aptn | grep PID).
• has security disadvantage: while your app (or any app that uses ports 443-1024) is down for some reason, another app could take the port. But this problem could also be applied to CAP/suid (in case you set it on interpreter, e.g. java/nodejs) and iptables-redirect. Use systemd-socket method to exclude this problem. Use authbind method to only allow special user binding.
• doesn't require setting CAP/suid every time you deploy new version of application.
• doesn't require application support/modification, like systemd-socket method.
• doesn't require kernel rebuild (if running version supports this sysctl setting)
• doesn't do LD_PRELOAD like authbind/privbind method, this could potentially affect performance, security, behavior (does it? haven't tested). In the rest authbind is really flexible and secure method.
• over-performs iptables REDIRECT/DNAT method, since it doesn't require address translation, connection state tracking, etc. This only noticeable on high-load systems.

Depending on the situation, I would choose between sysctl, CAP, authbind and iptables-redirect. And this is great that we have so many options.

You can setup a local SSH tunnel, eg if you want port 80 to hit your app bound to 3000:

sudo ssh $USERNAME@localhost -L 80:localhost:3000 -N

This has the advantage of working with script servers, and being very simple.

Modern Linux supports /sbin/sysctl -w net.ipv4.ip_unprivileged_port_start=0.

The standard way is to make them "setuid" so that they start up as root, and then they throw away that root privilege as soon as they've bound to the port but before they start accepting connections to it. You can see good examples of that in the source code for Apache and INN. I'm told that Lighttpd is another good example.

Another example is Postfix, which uses multiple daemons that communicate through pipes, and only one or two of them (which do very little except accept or emit bytes) run as root and the rest run at a lower privilege.

Bind port 8080 to 80 and open port 80:

sudo iptables -t nat -A OUTPUT -o lo -p tcp --dport 80 -j REDIRECT --to-port 8080
sudo iptables -A INPUT -p tcp --dport 80 -j ACCEPT

and then run program on port 8080 as a normal user.

you will then be able to access http://127.0.0.1 on port 80

TLDR: For "the answer" (as I see it), jump down to the >>TLDR<< part in this answer.

OK, I've figured it out (for real this time), the answer to this question, and this answer of mine is also a way of apologizing for promoting another answer (both here and on twitter) that I thought was "the best", but after trying it, discovered that I was mistaken about that. Learn from my mistake kids: don't promote something until you've actually tried it yourself!

Again, I reviewed all the answers here. I've tried some of them (and chose not to try others because I simply didn't like the solutions). I thought that the solution was to use systemd with its Capabilities= and CapabilitiesBindingSet= settings. After wrestling with this for some time, I discovered that this is not the solution because:

Capabilities are intended to restrict root processes!

As the OP wisely stated, it is always best to avoid that (for all your daemons if possible!).

You cannot use the Capabilities related options with User= and Group= in systemd unit files, because capabilities are ALWAYS reset when execev (or whatever the function is) is called. In other words, when systemd forks and drops its perms, the capabilities are reset. There is no way around this, and all that binding logic in the kernel is basic around uid=0, not capabilities. This means that it is unlikely that Capabilities will ever be the right answer to this question (at least any time soon). Incidentally, setcap, as others have mentioned, is not a solution. It didn't work for me, it doesn't work nicely with scripts, and those are reset anyways whenever the file changes.

In my meager defense, I did state (in the comment I've now deleted), that James' iptables suggestion (which the OP also mentions), was the "2nd best solution". :-P

>>TLDR<<

The solution is to combine systemd with on-the-fly iptables commands, like this (taken from DNSChain):

[Unit]
Description=dnschain
After=network.target
Wants=namecoin.service

[Service]
ExecStart=/usr/local/bin/dnschain
Environment=DNSCHAIN_SYSD_VER=0.0.1
PermissionsStartOnly=true
ExecStartPre=/sbin/sysctl -w net.ipv4.ip_forward=1
ExecStartPre=-/sbin/iptables -D INPUT -p udp --dport 5333 -j ACCEPT
ExecStartPre=-/sbin/iptables -t nat -D PREROUTING -p udp --dport 53 -j REDIRECT --to-ports 5333
ExecStartPre=/sbin/iptables -A INPUT -p udp --dport 5333 -j ACCEPT
ExecStartPre=/sbin/iptables -t nat -A PREROUTING -p udp --dport 53 -j REDIRECT --to-ports 5333
ExecStopPost=/sbin/iptables -D INPUT -p udp --dport 5333 -j ACCEPT
ExecStopPost=/sbin/iptables -t nat -D PREROUTING -p udp --dport 53 -j REDIRECT --to-ports 5333
User=dns
Group=dns
Restart=always
RestartSec=5
WorkingDirectory=/home/dns
PrivateTmp=true
NoNewPrivileges=true
ReadOnlyDirectories=/etc

# Unfortunately, capabilities are basically worthless because they're designed to restrict root daemons. Instead, we use iptables to listen on privileged ports.
# Capabilities=cap_net_bind_service+pei
# SecureBits=keep-caps

[Install]
WantedBy=multi-user.target

Here we accomplish the following:

• The daemon listens on 5333, but connections are successfully accepted on 53 thanks to iptables
• We can include the commands in the unit file itself, and thus we save people headaches. systemd cleans up the firewall rules for us, making sure to remove them when the daemon isn't running.
• We never run as root, and we make privilege escalation impossible (at least systemd claims to), supposedly even if the daemon is compromised and sets uid=0.

iptables is still, unfortunately, quite an ugly and difficult-to-use utility. If the daemon is listening on eth0:0 instead of eth0, for example, the commands are slightly different.