Hi Eliezer,
El 18/07/2014 15:03, Eliezer Croitoru escribió:
Hey Nicolas,
Can we change the way we are looking at the subject?
You first need to figure out the right way to send packets between the
"client" and the "server".
After this connection will be intact you will be able to configure the
other things such as interception the right way.
As a starter since the "server" has a public IP and the "client" has a
public IP it will be pretty easy to use a simple tunnel such as
ipip,sit,gre between the client and the server.
Then on the client side mark and then route traffic from the client to
the server without any DNAT\REDIRECT rules which is the best way to
handle this situation.
In your case TPROXY is not an option since the server cannot use the
client IP address for public use (probably due to routing policies
around the globe).
A good place to start is at:
http://www.linuxfoundation.org/collaborate/workgroups/networking/tunneling
http://tier.cs.berkeley.edu/drupal/howto/ip-tunnel-using-gre-on-linux
The above doesn't give you enough on basic routing which could be very
simple using "ip rule" and connection marking.
If you do not have a clue about it I will try to give you a real life
example I have used in the past to do this trick.
One of the examples I have used gre tunnel is for wccp interception
which should be similar to your setup but with a public IP addresses
classes mixed with local ones.
http://wiki.squid-cache.org/ConfigExamples/UbuntuTproxy4Wccp2
If you want I will be next evening(19 Jul) in the squid irc channel at
freenode:
http://webchat.freenode.net/?channels=squid
Eliezer
Thanks for the hints! I'm not sure whether I'll be able to try it before
you'll be on the Squid channel, but I'll let you know the results via
the list if not. I certainly have no experience with this subject, but
I'll check the documentation you provided and try to research over the
not detailed things.
However, I've made a much simpler experiment today. I setup a squid box
with the same characteristics as the remote, but inside the same network
as the client, so using private networking, and the result is exactly
the same... a forwarding loop. So having this, I guess I should
partially discard some public ip/different network/different
country/packet modifying issue by now and center on something much
simpler. This is for sure some overlooked detail making it enter the
loop, but I'm stuck at it for several days! I'll keep trying.
Thanks for the help!
On 07/18/2014 02:02 PM, Nicolás wrote:
Hi again,
El 18/07/2014 9:32, Amos Jeffries escribió:
On 18/07/2014 7:51 p.m., Nicolás wrote:
Hi Amos,
El 18/07/2014 5:21, Amos Jeffries escribió:
[...]
Unfortunately, this one neither seems to make a difference. On the
squid
box, the squid daemon is run by user proxy so I got the UID and
replaced
it in the rule you provided:
# id proxy
uid=13(proxy) guid=13(proxy) groups=13(proxy)
So on the client box:
iptables -t nat -A OUTPUT --match owner --uid-owner 13 -p tcp
--dport 80 -j ACCEPT
Then I appended the DNAT rule.
However, I should have (re)mentioned it: Additionally to have
different
public IP addresses, both the client box and the squid box are on
different networks (basically the squid box is located in a different
country than the client box). Should that make a difference?
Depends on how the packets get from one country to another. These are
the fundamental limits on packet handling:
1) There MUST NOT be any NAT manipulation of the packets destination IP
prior to their arrival on the Squid machine.
- However, source-IP:port manipulation as done by gateway machines
NATing outbound traffic from internal private source-IP to public
ranges
is okay.
2) a network interface tunnel should not be producing OUTPUT chain
packets. Traffic arriving through that type of tunnel should be dealt
with as per packets arriving on a standard NIC.
- Traffic arriving via application layer gateway (another proxy, or
userland VPN client) *might* go through the OUTPUT chain and need
handling as if sourced internally.
3) if the packets contain HTTP in port-80 syntax the "intercept" or
"tproxy" option is mandatory which one depending on method of capture.
- If the packets were destined *to another proxy* Squid does not
require the "intercept" option to receive.
Also if you wish, I can run squid with the -N -d options and send you
the output log, so you probably will rapidly know what else could be
happening there.
"debug_options 11,2" in squid.conf is going to provide you with a good
trace to see what is happening. Better than access.log will in these
circumstances.
Amos
I've made a trace to see what's happening and this is the result. I've
removed the duplicate requests for better readability. I've also
obfuscated the IPs to make them match the example above (A.B.C.D ->
client's public IP, E.F.G.H -> server's public IP).
2014/07/18 12:33:05| Starting Squid Cache version 3.3.8 for
x86_64-pc-linux-gnu...
2014/07/18 12:33:27.900| client_side.cc(2316) parseHttpRequest: HTTP
Client local=E.F.G.H:3128 remote=A.B.C.D:54341 FD 8 flags=33
2014/07/18 12:33:27.901| client_side.cc(2317) parseHttpRequest: HTTP
Client REQUEST:
---------
GET / HTTP/1.1
Host: www.google.es
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101
Firefox/24.0
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8
Accept-Language: es-ES,es;q=0.8,en-US;q=0.5,en;q=0.3
Accept-Encoding: gzip, deflate
Cookie:
PREF=ID=119a6e25e6eccb3b:U=95e37afd611b606e:FF=0:TM=1404500940:LM=1404513627:S=r7E-Xed2muOOp-ay;
NID=67=M5geOtyDtp5evLidOfam1uzfhl6likehxjXo7KcamK8c5jXptfx9zJc-5L7jhvYvnfTvtXYJ3yza7cE8fRq2x0iyVEHN9Pn2hz9urrC_Qt_xNH6IQCoT-3-eXTwb2h4f;
OGPC=5-25:
Connection: keep-alive
Pragma: no-cache
Cache-Control: no-cache
----------
2014/07/18 12:33:27.902| http.cc(2204) sendRequest: HTTP Server
local=E.F.G.H:43140 remote=E.F.G.H:3128 FD 11 flags=1
2014/07/18 12:33:27.902| http.cc(2205) sendRequest: HTTP Server REQUEST:
---------
[Same request with this line added:
Via: 1.1 FQDN (squid/3.3.8)]
----------
2014/07/18 12:33:27.902| client_side.cc(2316) parseHttpRequest: HTTP
Client local=E.F.G.H:3128 remote=E.F.G.H:43140 FD 13 flags=33
2014/07/18 12:33:27.902| client_side.cc(2317) parseHttpRequest: HTTP
Client REQUEST:
---------
[Same request]
----------
2014/07/18 12:33:27.903| client_side.cc(1377) sendStartOfMessage: HTTP
Client local=E.F.G.H:3128 remote=E.F.G.H:43140 FD 13 flags=33
2014/07/18 12:33:27.903| client_side.cc(1378) sendStartOfMessage: HTTP
Client REPLY:
---------
HTTP/1.1 403 Forbidden
Server: squid/3.3.8
Mime-Version: 1.0
Date: Fri, 18 Jul 2014 10:33:27 GMT
Content-Type: text/html
Content-Length: 3932
X-Squid-Error: ERR_ACCESS_DENIED 0
Vary: Accept-Language
Content-Language: es-es
X-Cache: MISS from ovh.devels.es
X-Cache-Lookup: MISS from ovh.devels.es:3127
Via: 1.1 ovh.devels.es (squid/3.3.8)
Connection: keep-alive
----------
2014/07/18 12:33:27.903| ctx: enter level 0: 'http://www.google.es/'
2014/07/18 12:33:27.903| http.cc(761) processReplyHeader: HTTP Server
local=E.F.G.H:43140 remote=E.F.G.H:3128 FD 11 flags=1
2014/07/18 12:33:27.903| http.cc(762) processReplyHeader: HTTP Server
REPLY:
---------
[HTTP error page saying "Access denied"]
From what I see, squid makes several times requests to itself instead
of redirecting the request outside. Shouldn't there be a rule
redirecting that traffic to the internet? On the server side I have no
such rule right now.
