In the old days, we used telnet to see if a port on a remote host was open: telnet hostname port would attempt to connect to any port on any host and give you access to the raw TCP stream.

These days, the systems I work on do not have telnet installed (for security reasons), and all outbound connections to all hosts are blocked by default. Over time, it's easy to lose track of which ports are open to which hosts.

Is there another way to test if a port on a remote system is open – using a Linux system with a limited number of packages installed, and telnet is not available?

  • Related: check status of one port on remote host at SO
    – kenorb
    Dec 30 '15 at 14:50
  • I was having this same issue. The answer by @Subhranath Chunder below helped. However, I then found out that installing Telnet was a small matter of running brew install telnet. So I expect Linux users can do the same with yum and apt-get.
    – Mig82
    Sep 26 '19 at 9:22
  • When "all outbound connections to all hosts are blocked by default" there will be no way to perform such a test - you are offline Nov 27 '20 at 4:37

14 Answers 14


Bash has been able to access TCP and UDP ports for a while. From the man page:

    If host is a valid hostname or Internet address, and port is an integer port number
    or service name, bash attempts to open a TCP connection to the corresponding socket.
    If host is a valid hostname or Internet address, and port is an integer port number
    or service name, bash attempts to open a UDP connection to the corresponding socket.

So you could use something like this:

xenon-lornix:~> cat < /dev/tcp/
SSH-2.0-OpenSSH_6.2p2 Debian-6
^C pressed here

Taa Daa!

  • 1
    However on ports that were not open it timed out after 22 seconds (tried on Ubuntu 14.04 (Trusty Tahr) for a remote server). Interestingly, the timeout period is much shorter than the one for nc (see thnee's answer). Jul 21 '15 at 13:00
  • 2
    @lornix, ok, but in this case I have to get the same result with use nc without -z option, but it still does not work: # nc -v -w5 18080 Connection to 18080 port [tcp/*] succeeded! # cat < /dev/tcp/ Just hangs without any result. Just want to understand when I can use "/dev/tcp/host/port" option
    – Alexandr
    Jun 14 '16 at 7:09
  • 6
    @Alexandr... actually, "hangs without any result" is pretty much expected behavior. cat is waiting for input. nc has extra smarts to enable it to sense no-data pending and stops trying. cat isn't quite as smart. Try cat < /dev/tcp/localhost/22, you should get your sshd header. Evidently, your process on port 18080 waits for something to come in, before sending anything. Port 22 (ssh) greets you with it's version and whatnot. Try it out!
    – lornix
    Jun 14 '16 at 8:44
  • 1
    @lornix, thank you very much for explanation! Now the restriction is clear. I think using nc should be a preferred way to check ports.
    – Alexandr
    Jun 14 '16 at 14:29
  • 5
    This was incredibly helpful when working with a docker container that had nothing installed. Was able to quickly verify that the container had access to non-containerized DB via DNS. Example: cat < /dev/tcp/hostname/5432 Oct 24 '18 at 14:09

Nice and verbose! From the man pages.
Single port:

nc -zv 80

Multiple ports:

nc -zv 22 80 8080

Range of ports:

nc -zv 20-30
  • 2
    Perfect! This outputs a clear connection succeeded/failed message with just the one line. Note that multiple ranges don't appear to work for my nc version; only the first range is tested.
    – davidjb
    May 18 '15 at 3:42
  • 7
    This hanged when tried on Ubuntu 14.04 (Trusty Tahr) for a remote server (same LAN) for closed ports (it timed out after 127 seconds) - thus not very suitable in scripts. It did work though for a service that had a port open. Using option "-w2" could be the solution. Jul 21 '15 at 12:36
  • 8
    Use -u option for UDP ports.
    – Efren
    Sep 21 '16 at 0:10
  • 9
    On version 6.4 of ncat -z is not recognized. I was able to do without z
    – smishra
    Apr 6 '17 at 17:18
  • 6
    You can check multiple ranges with: nc -zv 22,80,8080,20-30,443-446 (nc Version: 1.107-4).
    – bobbel
    Jul 6 '17 at 16:55

Netcat is a useful tool:

nc 123 &> /dev/null; echo $?

Will output 0 if port 123 is open, and 1 if it's closed.

  • 2
    This is a far more elegant and scriptable answer than my own. It is unfortunate for me that the security-conscious sysadmins who withheld telnet also withheld nc (though – strangely – not curl or wget).
    – Steve HHH
    Jul 19 '13 at 19:51
  • 1
    Yes that is completely arbitrary and silly.
    – thnee
    Jul 19 '13 at 20:10
  • 3
    Let the FOR statements begin! Jul 19 '13 at 21:37
  • 2
    This hanged when tried on Ubuntu 14.04 (Trusty Tahr) for a remote server (same LAN) for closed ports (it timed out after about 127 seconds) - thus not very suitable in scripts. It did work though for a service that had a port open, returning 0. Using option "-w2" could be the solution. Jul 21 '15 at 13:06
  • 1
    I think -G 2 would be more appropriate for TCP timeout
    – A B
    Aug 24 '15 at 21:45

The simplest method, without making use of another tool, such as socat, is as described in @lornix's answer above. This is just to add an actual example of how one would make use of the psuedo-device /dev/tcp/... within Bash if you wanted to, say, test if another server had a given port accessible via the command line.


Say I have a host on my network named skinner.

$ (echo > /dev/tcp/skinner/22) >/dev/null 2>&1 \
    && echo "It's up" || echo "It's down"
It's up

$ (echo > /dev/tcp/skinner/222) >/dev/null 2>&1 && \
    echo "It's up" || echo "It's down"
It's down

The reason you want to wrap the echo > /dev/... in parentheses like this, (echo > /dev/...) is because if you don't, then with tests of connections that are down, you'll get these types of messages showing up.

$ (echo > /dev/tcp/skinner/223) && echo hi
bash: connect: Connection refused
bash: /dev/tcp/skinner/223: Connection refused

These can't simply be redirected to /dev/null since they're coming from the attempt to write out data to the device /dev/tcp. So we capture all that output within a sub-command, i.e. (...cmds...) and redirect the output of the sub-command.

  • 1
    This is excellent. Wish it would get voted up to the top. I only read this far down the page because I accidentally scrolled before closing it.
    – Still.Tony
    Feb 12 '15 at 13:44
  • @Okuma.Tony - yes that's always an issue with Q's that have many answers 8-). Thanks for the feedback though, it's appreciated.
    – slm
    Feb 12 '15 at 14:26
  • I like it. I made a script out of it for my own use!
    – Marinaio
    Nov 9 '21 at 15:41

I found that curl may get the job done in a similar way to telnet, and curl will even tell you which protocol the listener expects.

Construct an HTTP URI from the hostname and port as the first argument to curl. If curl can connect, it will report a protocol mismatch and exit (if the listener isn't a web service). If curl cannot connect, it will time out.

For example, port 5672 on host is either closed or blocked by a firewall:

$ curl
curl: (7) couldn't connect to host

However, from a different system, port 5672 on host can be reached, and appears to be running an AMQP listener.

$ curl
curl: (56) Failure when receiving data from the peer

It's important to distinguish between the different messages: the first failure was because curl could not connect to the port. The second failure is a success test, though curl expected an HTTP listener instead of an AMQP listener.

  • 7
    If curl isn't available, wget might be. wget -qS -O- http://ip.add.re.ss:port should effectively do the same thing.
    – user
    Jul 19 '13 at 17:52
  • 4
    This even works with a hostname, ex. curl myhost:22.
    – 에이바
    Feb 25 '14 at 16:17
  • This may be incorrect. I am havng a tomcat service running, but getting 404 error. # curl -k <html><head><title>Apache Tomcat/7.0.34 - Error report</title><style><!--H1 --- HR {color : #525D76;}--></style> </head><body><h1>HTTP Status 404 - /</h1><HR size="1" noshade="noshade"><p><b>type</b> Status report</p><p><b>message</b> <u>/</u></p><p><b>description</b> <u>The requested resource is not available.</u></p><HR size="1" noshade="noshade"><h3>Apache Tomcat/7.0.34</h3></body></html> Jun 2 '15 at 6:30
  • See my post with similar approach.
    – kenorb
    Dec 30 '15 at 14:49
[admin@automation-server 1.2.2]# nc -v -z -w2 6443
nc: connect to port 6443 (tcp) failed: Connection refused

[admin@automation-server 1.2.2]# nc -v -z -w2 6443
Connection to 6443 port [tcp/sun-sr-https] succeeded!

Hope it solves your problem :)

  • 1
    Yes, this is better - timing out almost immediately for closed ports. Jul 21 '15 at 12:38
  • 1
    Does this always use TCP or is there a way to get it to check UDP?
    – kmoe
    Jan 2 '16 at 17:09

Here is one-liner:

</dev/tcp/localhost/11211 && echo Port is open || echo Port is closed

using Bash syntax explained in @lornix answer.

For more info, check: Advanced Bash-Scripting Guide: Chapter 29. /dev and /proc.

  • 1
    what is the timeout for closed ports?
    – Tilo
    Dec 12 '18 at 17:42

Combining the answers from @kenorb and @Azukikuru you could test port open/closed/firewalled.

timeout 1 bash -c '</dev/tcp/ && echo Port is open || echo Port is closed' || echo Connection timeout

Another approach with curl for reaching any port

curl telnet://

I was struggling for a whole day because none of these answers seemed to work for me. The problem is that the most recent version of nc no longer has the -z flag, whereas direct access via TCP (as according to @lornix and @slm) fails when the host is not reachable. I eventually found this page, where I finally found not one but two working examples:

  1. nc -w1 22 </dev/null

    (the -w flag takes care of the timeout, and the </dev/null replaces the -z flag)

  2. timeout 1 bash -c '(echo > /dev/tcp/ >/dev/null 2>&1'

    (the timeout command takes care of the timeout, and the rest is from @slm)

Then, simply use && and/or || (or even $?) to extract the result. Hopefully, somebody will find this information useful.


Here's a function that will pick one of the methods depending on what's installed on your system:

# Check_port <address> <port> 
check_port() {
if [ "$(which nc)" != "" ]; then 
elif [ "$(which curl)" != "" ]; then
elif [ "$(which telnet)" != "" ]; then
elif [ -e /dev/tcp ]; then
      if [ "$(which gtimeout)" != "" ]; then  
      elif [ "$(which timeout)" != "" ]; then  
echo "Using $tool to test access to $1:$2"
case $tool in
nc) nc -v -G 5 -z -w2 $1 $2 ;;
curl) curl --connect-timeout 10 http://$1:$2 ;;
telnet) telnet $1 $2 ;;
gtimeout)  gtimeout 1 bash -c "</dev/tcp/${1}/${2} && echo Port is open || echo Port is closed" || echo Connection timeout ;;
timeout)  timeout 1 bash -c "</dev/tcp/${1}/${2} && echo Port is open || echo Port is closed" || echo Connection timeout ;;
devtcp)  </dev/tcp/${1}/${2} && echo Port is open || echo Port is closed ;;
*) echo "no tools available to test $1 port $2";;

export check_port
  • why would you set tool, only to use it later in the case statement. isn't it simpler to check which tool is available and use it immediately in the if block? Nov 5 '19 at 15:46
  • Good point -- I threw that together in a hurry -- it does make sense to collapse the code as you suggested. On the other hand, setting tool could be useful if there's a need to do more with it in another part of the code. May 14 '20 at 15:05

It shouldn't be available on your box, but try with nmap.

  • 4
    nmap is a good tool, but not available on these systems. Rather than download nmap, compile it, install it to my home directory, then copy it to all the other systems, I was hoping to find a way using existing tools available in most Linux installations.
    – Steve HHH
    Jul 19 '13 at 17:16

for reference, expanding on @peperunas' answer:

the way to use nmap to test, is:

nmap -p 22

(example above uses localhost for demonstration purposes)


If you have curl installed:

curl -v telnet://$host:$port/$path
  • (1) I presume that you mean that the curl -v telnet://, the : and the (final) / are literal, and $host and $port are placeholders for the name/address of the host in question and the port number in question.  What should the user use for $path?  (2) If the port is open, but it is not implementing HTTP, we can expect this command to fail.  So how does one distinguish between a failure because the port isn’t open and a failure because the port isn’t HTTP? … (Cont’d)
    – Scott
    Oct 14 '20 at 21:13
  • (Cont’d) … (3) curl http://host:port has already been given as an answer (which was clear on point #1 and discussed point #2).  Are you saying that curl -v telnet: is superior to curl http:  Why?  (4) And there’s another answer that suggests curl http://host:port, and there’s even another one suggesting curl telnet://host:port.  What does your answer add to those earlier ones? … … … … … … … … … … … … … … … … … … … … … Please do not respond in comments; edit your answer to make it clearer and more complete.
    – Scott
    Oct 14 '20 at 21:13

If you've to test more than on system you may use our test tool dda-serverspec (https://github.com/DomainDrivenArchitecture/dda-serverspec-crate) for such tasks. You may define your expectation

{:netcat [{:host "mywebserver.com" :port "443"}
          {:host "telnet mywebserver.com" :port "80"}
          {:host "telnet mywebserver.com" :port "8443"}]}

and test these expectation either against localhost or against remote hosts (connect by ssh). For remote tests you've to define a targets:

{:existing [{:node-name "test-vm1"
             :node-ip ""}
            {:node-name "test-vm2"
             :node-ip ""}]
 :provisioning-user {:login "ubuntu"}}

You may run the test with java -jar dda-serverspec.jar --targets targets.edn serverspec.edn

Under the hood we're using netcat as proprosed above ...

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.