问题原文
原帖链接 · mohrphium · CC BY-SA 3.0
I have HAProxy for my two sites, one of them public and one private.
www.mysite.com private.mysite.com
Atm, I'm using haproxy like this:
frontend mysite_https
bind *.443 ssl crt /etc/mycert.pem ca-file /etc/myca.pem verify optional no-sslv3
mode http
acl domain_www hdr_beg(host) -i www.
acl domain_private hdr_beg(host) -i private.
acl path_ghost path_beg /ghost/
acl clientcert ssl_c_used
redirect location https://www.example.com if path_ghost !clientcert
redirect location https://www.example.com if !domain_www !clientcert
use_backend bknd_private if domain_private
use_backend bknd_www if domain_www
default_backend bknd_www
What this should do is ask for a client certificate (optionally) and proceed. If the domain is not www.example.com and the visitor cannot provide the right certificate or the path is /ghost/ and the visitor cannot provide the right certificate, it should be redirected to https://www.example.com
So far, this works fine. However, I got complaints by Mac users browsing my site with Safari that they keep getting asked for the certificate when there browsing on https://www.example.com/ whereas for example Firefox only asks when browsing https://private.example.com/ or https://www.example.com/ghost/.
Appearently that's just how Safari works so I can't fix that. My idea was to use SNI to divide between different frontends
frontend mysite_https
bind *.443 ssl crt /etc/mycert.pem no-sslv3
frontend private_https
bind *.443 ssl crt /etc/mycert.pem ca-file /etc/myca.pem verify optional no-sslv3
Of course that doesn't work because
a. I can't have two frontends listening on port 443 with only one public IP b. I haven't found a way yet to say "use_frontend if domain_www" or something like that. (Only use_backend or use-server)
I also tried doing it with three haproxy servers
frontend haproxy-sni
bind *:443 ssl crt /etc/mycert.pem no-sslv3
mode tcp
tcp-request inspect-delay 5s
tcp-request content accept if { req.ssl_hello_type 1 }
acl domain_www ssl_fc_sni_end -i www.example.com
use-server server1 haproxy-private.lan if !domain_www
use-server server2 haproxy-public.lan if domain_www
This works, the problem here however is that haproxy-private asks for the client certificate, but the request doesn't reach the browser. Somehow haproxy-sni drops the request.
Also, I now have three haproxy servers which is not desirable (although a possible option if I can't find a better solution).
Preferably I would like something like this (made up.. don't know the real options)
frontend mysite_https
bind *.443 ssl crt /etc/mycert.pem no-sslv3
mode http
acl domain_www hdr_beg(host) -i www.
acl domain_private hdr_beg(host) -i private.
acl path_ghost path_beg /ghost/
ssl_options ca-file /etc/myca.pem verify optional if !www_domain # made up!
ssl_options ca-file /etc/myca.pem verify optional if !path_ghost # made up!
acl clientcert ssl_c_used
redirect location https://www.example.com if path_ghost !clientcert
redirect location https://www.example.com if !domain_www !clientcert
...
I hope someone can help me with this...
已采纳解答
原帖链接 · mohrphium · CC BY-SA 3.0
I found a solution to this problem, that doesn't require additional servers or services. I'm not entirely sure if this doesn't spawn new problems though. For me it seems to work right now.
The way I did it, was to create a frontend for each domain that required different ssl settings. I then set the bind option of those frontends to high ports (these are not reachable from public!).
I created another frontend listening on port :443 to divide traffic based on SNI, and set the backend servers to 127.0.0.1:high-port.
This way, I created sort of a loop in haproxy
[incoming]->[haproxy:443]->[haproxy:7000]->[www.intern.lan]
[incoming]->[haproxy:443]->[haproxy:8000]->[private.intern.lan]
Here is the config part.
frontend frnd_snipt # Frontend_SNI-PassThrough (snipt)
bind *:443 # Do not use bind *:8443 ssl crt etc....!
option tcplog
mode tcp
tcp-request inspect-delay 5s
tcp-request content accept if { req_ssl_hello_type 1 }
acl subdomain_is_www req_ssl_sni -i www.example.com
acl subdomain_is_www req_ssl_sni -i example.com
acl subdomain_is_private req_ssl_sni -i private.example.com
use_backend bknd_snipt_private if subdomain_is_private
use_backend bknd_snipt_www if subdomain_is_www
backend bknd_snipt_www
mode tcp # tcp mode must match the frontend mode - already set as default in [global]
server snipt-www 127.0.0.1:7000 # run without "check", otherwise haproxy checks itself all the time!
backend bknd_snipt_private
mode tcp
server snipt-private 127.0.0.1:8000 # also, don't add "ssl" when in tcp mode. "ssl" is an http mode option (result in "NO-SRV" when set in tcp)
##### NORMAL HAPROXY PART #####
frontend www_example_com # this frontend can be in tcp or http mode...
bind *:7000 ssl crt /etc/mycert.pem no-sslv3 # www. frontend with normal https
mode http
option httplog
frontend private_example_com
bind *:8000 ssl crt /etc/mycert.pem ca-file /etc/myca.pem verify optional no-sslv3 # private. frontend with client certificate request.
mode http
option httplog
... # whatever you have in your frontend
If anyone has thoughts on this, or any idea why this could be a bad idea please let me know. It works, but I'm wondering why use_frontend isn't an option. Maybe because it's something that shouldn't be done for whatever reasons.
摘自公开归档,保留原文语言,版式经整理,外部媒体与邮箱地址已移除。
