On 7/26/23 13:58, Mark Thomas wrote:
I'm not a huge fan of this feature in general. I prefer supporting features backed by specifications rather than vendor specific hacks.

I think the PROXY protocol is fairly standard, even if it's not backed by an RFC. It's published by haproxy, but supported by nginx, (obviously) haproxy, AWS, httpd[1], and a whole bunch of others (

Well, the reality is that people want to use this in the real world and this is essentially the only way to do it, barring coming up with a whole new protocol for the purpose (I'm looking at /you/ AJP!).

So why not use /the/ protocol that (a) exists and (b) is supported by every single product that currently supports this type of thing?

My support for any patch is going to depend on the specifics of the patch.

In addition to the comments in the BZ
- exposing the data as a request attribute is inconsistent with other
   mechanisms that solve the same problem (e.g. see RemoteIpFilter)


The whole point of PROXY is to kind of mix-together the capabilities of both the RemoteIPFilter/Valve (which uses HTTP headers for source-information) and the top-level idea of a Connector (something that binds to a socket and pushes bytes around).

The confusing thing here is that those two jobs are performed at relatively different levels in Tomcat at the moment, as I understand things.

If some kind of UberConnector could be built which essentially does something like the following, it would be ideal:

public void accept(Socket s) {
  ProxyHeader proxyHeader = readProxyHeader(s);

  Connector realConnector = getRealConnector();



I'm sure there are other pieces of information that would be good to pass-through, but the identity of the remote client is the most interesting one.

- needs to be implemented for all Connectors

I hope not. The connectors should be able to just have a thin layer in front of them "sipping" the header off the beginning of the connection. I am *way* out of my depth here when it comes to Tomcat internals and so I don't want to appear to be telling you (Mark) "how it works/should work", but conceptually it "seems easy". That may not translate into "easy implementation" or it may mean "tons of refactoring that we wouldn't need if we didn't care that much."

- I'd expect it to look more like the SNI processing

SNI processing is very connector-dependent, of course, because it's HTTPS-only. PROXY should allow HTTP, HTTPS, AJP, SFTP, JDBC, anything. So if it can be implemented as something that can just "sit in front of" *any* connector now or in the future of Tomcat, that would be ideal. It could definitely be implemented as an "optional feature" on a Connector-by-Connector basis, but my sense is that it can be done separately and globally.

Again, I'm speaking from a position of profound ignorance, here. Please don't hear me say "oh, this is easy, Mark... just go do it!" :)

Generally, I don't think implementing this is going to be possible as some sort of plug-in.

+1 Unless the plug-in is "a whole new set of protocol/endpoint/etc. handlers" which is a rather serious commitment.


[1] search for "haproxy"

On 26/07/2023 17:44, Amit Pande wrote:
Missed to ask this:

Looking the patch, it involves modifying Tomcat code.
Was wondering if it would be possible to refactor this patch and/or allow Tomcat core code to extend and plug-in the proxy protocol support?


-----Original Message-----
From: Amit Pande
Sent: Wednesday, July 26, 2023 11:43 AM
To: Tomcat Users List <>
Subject: RE: [External] Re: Supporting Proxy Protocol in Tomcat

Chris, Mark,

Any thoughts on this?

Mark, if we clean up the patch and re-submit, do you will have any concerns (specially security wise)?


-----Original Message-----
From: Jonathan S. Fisher <>
Sent: Monday, July 24, 2023 12:41 PM
To: Tomcat Users List <>
Subject: Re: [External] Re: Supporting Proxy Protocol in Tomcat

Just a side note, because we're also very interested in this patch!

Awhile back, I was successfully able to apply this patch and terminate TCP/TLS using HaProxy. We then had Tomcat listen on a unix domain socket and the Proxy protocol provided *most *of the relevant/required information to tomcat. I believe we had to add a Valve to tomcat to set the Remote IP however as the patch didn't handle that case.

I can find my notes from that experiment, but I do remember getting a significant boost in throughput and decrease in latency.

+1 for this patch and willing to help out!

On Mon, Jul 24, 2023 at 11:22 AM Amit Pande <>

Thank you, Chris, again for inputs.
And sorry to circle back on this, late.

One related question is - does it make sense to use the patch attached
in ?
And potentially, get it integrated into Tomcat versions?

There are concerns from Mark about using the patch in its current
state, but I see last comment (#24) on the issue and looks like there
are some more points to be concluded.


-----Original Message-----
From: Christopher Schultz <>
Sent: Wednesday, May 10, 2023 4:21 PM
Subject: Re: [External] Re: Supporting Proxy Protocol in Tomcat


On 5/10/23 12:59, Amit Pande wrote:
Yes, we intended to have Tomcat run behind a (transparent) TCP proxy e.g.

which supports the proxy protocol.

Since there is not much action on this
 does it imply that most of the times Tomcat is running behind HTTP proxies and not TCP 
Or does it mean that, Tomcat or applications running in Tomcat does
need the remote client address information?

I can't speak for anybody else, but I use Apache httpd as my
reverse-proxy and I do terminate TLS. I also use it for
load-balancing/fail-over, caching, some authorization, etc. I wouldn't
be able to use a TCP load-balancer because I hide multiple services
behind my reverse-proxy which run in different places. It's not just s dumb pass-through.

Hope that helps,

-----Original Message-----
From: Christopher Schultz <>
Sent: Monday, May 8, 2023 3:40 PM
Subject: [External] Re: Supporting Proxy Protocol in Tomcat


On 5/4/23 16:07, Amit Pande wrote:
We have a similar requirement as mentioned in the below enhancement


Is there any plan to add this support in Tomcat in future releases?

Nothing at the moment that I know of.

I thought that markt had looked at this a while back and said it
look too difficult. It does require Tomcat to handle the stream
directly and not just rely on Java's SSLServerSocket. I thought that
had been done at some point, but it may not have. Handling the stream
directly may have some other advantages as well, though it definitely
makes the code more complicated.

Also, since this was requested long time back and there is no
update, are there any other alternatives to pass the client
information from load balancer to Tomcat in situations where there
is no SSL termination at load balancer?
You mean like a network load balancer where the lb is just proxying
bytes and not looking at the data at all? The PROXY protocol really is
the best way to do that, honestly.


- To unsubscribe, e-mail:
For additional commands, e-mail:

- To unsubscribe, e-mail:
For additional commands, e-mail:

To unsubscribe, e-mail:
For additional commands, e-mail:

To unsubscribe, e-mail:
For additional commands, e-mail:

Jonathan |
Pessimists, see a jar as half empty. Optimists, in contrast, see it as half full. Engineers, of course, understand the glass is twice as big as it needs to be.

To unsubscribe, e-mail:
For additional commands, e-mail:

To unsubscribe, e-mail:
For additional commands, e-mail:

To unsubscribe, e-mail:
For additional commands, e-mail:

Reply via email to