Forum Discussion

Piotr_Lewandows's avatar
Piotr_Lewandows
Icon for Altostratus rankAltostratus
Nov 20, 2017

URI Rule in Rewrite profile disable Server SSL profile?

Hi,

 

Setup:

 

  • TMOS 12.1.2HF1
  • VS listening on port 80
  • Server SSL profile attached
  • Rewrite profile settings:
  • Pool member set to port 443

Whith above settings SSL is disabled on backend - client HTTP request on port 80 is passed as HTTP request on port 443 on backend side. There is no SSL Handshake performed.

 

Result: backend server do not respond

 

If URI Rule is changed to:

 

Everything starts to work, BIG-IP is starting SSL Handshake and backend is accepting connection.

 

Is that by design? So protocol definition in URI Rule is defining protocol used on backend - in other words disables Server SSL profile when set to http for Server?

 

Piotr

 

9 Replies

  • Piotr,

     

    I guess you discovered a feature not documented.

     

    you can :

     

    • contact your local F5 team to ask for information.
    • open a support case to confirm it's the expected behavior
    • troubleshoot yourself how it is working (multiple rewrite rules depending on the URI and capture server side traffic to see if the F5 send the CLIENT_HELLO packet)

    Of course, don't forget to share here the information.

     

    I guess the last solution will be quicker than others ;-) . we are still waiting for a clientless feature documentation since several years.

     

  • Hi,

     

    Will try to go via Support.

     

    I am not sure about your last advice:

     

    troubleshoot yourself how it is working (multiple rewrite rules depending on the URI and capture server side traffic to see if the F5 send the CLIENT_HELLO packet)

     

    I already checked that when http is set in Server URI then BIG-IP is not sending any CLIENT_HELLO packets but still sending to port 443.

     

    Immediately after changing http to https Server SSL profile starts to work - normal SSL Handshake performed and traffic encrypted.

     

    What is strange that seems to be not a case if Local Traffic Policy is used to switch between Pools with selective disabling Server SSL.

     

    Will have to test but I pretty sure that even if there is URI Rule http <-> http but selected pool member is using Server SSL it is not disabled - encryption works.

     

    Piotr

     

  • I confirm that even if Local Traffic Policy is configured to disable serverssl, it first send CLIENT_HELLO, then fallback to cleartext...

     

    one customer opened a support case and the answer was to disable serverssl on server side irule event.

     

  • Hmm, do you mean that disabling Server SSL via LTP is not working? I had LTP where depending on request Host header Server SSL is kept enabled (two FQDNS) and disabled (one FQDN) - never noticed issue with sending unecrypted traffic to backend - but never noticed CLIENT_HELLO as well because most often my Wireshark was set with http filter :-)

     

    I did however notice some strange errors in LTM log related to VS with LTP:

     

    transaction failed:010716e2:3: Policy '/Common/Drafts/something', rule 'some rule'; an action precedes its conditions.

     

    This is in fact related to rule that is not disabling Server SSL.

     

    Piotr

     

  • we discovered this behavior after upgrade to version 13.0 because of ssl log level change since version 12.0 https://support.f5.com/csp/article/K15292

    the issue is a serverssl disable / enable is per client side connection on client side context. if the first server side connection used a SSL connection, it can't disable SSL for the following server side connection.

    to summarize, SSL::disable serverside or LTP disable serverssl can't disable server side ssl per request.

    I wrote following rule for this behavior (this is not the one I tested, I rewrite it here)

    when HTTP_REQUEST {
        if {[POLICY::targets serverssl] } {
            set disable_server_ssl 1
        } else {
            set disable_server_ssl 0
        }
    }
    
    when SERVER_CONNECTED {
        if { $disable_server_ssl == 1 } then {
            SSL::disable serverside
        }
    }
    
  • Thanks for info. To be 100% sure - is this related to v13 or it should not work as well in v12?

     

    I will redo the tests but I am 99% sure that I tried such sequence (one VS with LTP and Server SSL profile):

     

    • Connect to FQDN directing to SSL backend - works
    • Connect to FQDN directing to no SSL backend - works
    • Connect to FQDN directing to another SSL backend - works

    Never noticed this issue - but I am on 12.1.2 as already mentioned.

     

    Piotr

     

  • Thanks for info, good to know. I never used such scenario so that is why I never experienced problem.

     

    I thing that it can be caused by fact that same TCP Connection on client side is used for request to different URIs that can be passed to different backends - some with SSL some without - what do you think?

     

    Wonder if OneConnect or some one to force client to establish new TCP connection when there is backend switch would solve it?

     

    Piotr

     

  • Hi Stan,

    Weird that I forgot about this conversation 🙂

    One correction to your iRule:

    if {[POLICY::targets serverssl] } {
    should be
    if {[POLICY::targets server-ssl] } {

    BTW - thanks for pointing out how to use POLICY::targets, it is very useful to make iRule more universal!

    Did some tests and can't find a problem with server ssl disable in LTP, either in 13.1.1.3 it was corrected or I have not enough sample traffic to trigger error.

    Tested LTP like below switching between HTTPS and HTTP to the same server based on URL. It just works. / is sending traffic over HTTPS and /nossl over HTTP, I can repeat switching forever and can't trigger error.

    LTP used

    ltm policy test-disable-serverssl-per-request {
        controls { forwarding server-ssl }
        description "If URL is /nossl disable serverssl"
        draft-copy Drafts/test-disable-serverssl-per-request
        last-modified 2019-02-28:11:50:07
        requires { http }
        rules {
            change-host_r {
                actions {
                    0 {
                        log
                        write
                        facility local0
                        message "tcl:VS: \"[getfield [virtual name] / 3]\"; Host: \"[HTTP::host]\"; URI: \"[HTTP::uri]\""
                        priority info
                    }
                    1 {
                        http-host
                        replace
                        value "tcl:[class match -value [getfield [HTTP::host] : 1] equals vh-parent-https-https-vs_dg]"
                    }
                }
                description "Change Host header for any other URL"
                ordinal 1
            }
            disable-serverssl-via-url_r {
                actions {
                    0 {
                        log
                        write
                        facility local0
                        message "tcl:VS: \"[getfield [virtual name] / 3]\"; Host: \"[HTTP::host]\"; URI: \"[HTTP::uri]\""
                        priority info
                    }
                    1 {
                        http-host
                        replace
                        value "tcl:[class match -value [getfield [HTTP::host] : 1] equals vh-parent-https-https-vs_dg]"
                    }
                    2 {
                        server-ssl
                        disable
                    }
                    3 {
                        forward
                        select
                        pool vh-site41-80_pl
                    }
                }
                conditions {
                    0 {
                        http-uri
                        path-segment
                        index 1
                        values { nossl }
                    }
                }
                description "Disables serverssl when /nossl URL is discovered; changes Host header"
            }
        }
        status published
        strategy first-match
    }
    

    So I am bit puzzled how to trigger issue you described - any hint?

    Piotr