Re: [Pound Mailing List] POODLE, Pound and Branches

Joe-

Your 2.6pcidss branch works well.  Unfortunately 2.6pcidss does not have a
resolution for http://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2005-2090.
Pound, understandably, recalculates the content-length header and sends it
to the back end, effectively masking the multiple content-length headers
submitted by the would be hacker.  There really is nothing wrong with this
other than we continue to fail PCI scans from Qualys.  I assume other
people are passing their scans by registering a false positive with the
scanner.

I'm attaching a patch to show what we've done.



On Thu, Oct 23, 2014 at 2:59 PM, Joe Gooch <[email protected]> wrote:

> There seems to be some confusion, so I'm going to go through all of this
> again.
>
> Robert maintains the mainline Pound branch.  2.7d is the most recent
> release, and it's beta quality.  There will be no mainline changes to
> the 2.6 branch.  (Robert might make a different decision, but this would
> be the first time in recent history that that's ever happened)
>
> I maintain two branches of Pound.  One is a branch I maintain with the
> patches I'd like to see in the next 2.7 version... at this point, 2.7e.
> The other is called pcidss/v2.6, which is Pound 2.6, plus cipher and
> protocol patches necessary (initially) to pass PCI compliance, and as
> part of that is the directive to disable SSL3.
>
> This branch is on github here:
> https://github.com/goochjj/pound/tree/pcidss/v2.6
>
> with a Zip here:
> https://github.com/goochjj/pound/archive/pcidss/v2.6.zip
>
> I know a lot of people are using this branch.  I hear back from them
> every once in awhile, and I know they're on the list.  If you don't want
> to run beta code, but you still want the cipher and protocol patches,
> you must run this branch. Which is *not* an official release from
> Robert, apsis, or any other official entity.  All source code is
> available on github, you can see every change I've made and for the most
> part, the patches have already been posted to the mailing list.  I don't
> warrant this in any official matter, I just provide it as a service to
> the community.
>
> You could assemble these patches yourself, if you so choose.  That's up
> to you.
>
> Pound has had Cipher designation support for quite some time, but you
> cannot *solely* mitigate POODLE using the ciphers line.  Some of the
> ciphers available in TLSv1.0 and later are *also* available in SSLv3,
> and if you tell a server not to use a cipher valid for SSLv3, it either
> 1) won't use that cipher at all, which means it's not available for TLS,
> which is wrong... or 2) it'll see it's available for TLS and re-enable
> it for SSLv3 later in the cipher string.  (i.e. -SSLv3:+TLSv1)
>
> Either way this isn't what you want.
>
> That's why Apache, nginx, and now pound have a separate directive where
> you can disable SSL protocols.
>
> In the pcidss/v2.6 branch, you do so with:
>   DisableSSLv2
>   DisableSSLv3
>
> If you're running the official 2.7d (or higher) branch, you do so with:
>   Disable SSLv3
>
> which also implicitly disables SSLv2.
>
> To get an A on SSLLabs, you'll need to do *more* than that.  I run this:
> 8<----------------
> #dh2048.pem generated with openssl dhparams -5 2048 -out dh2048.pem
> DHParams        "dh2048.pem"
> ECDHCurve       prime256v1
>
> ListenHTTPS
>    .....
>         SSLAllowClientRenegotiation     0
>         SSLHonorCipherOrder     1
>
>         Ciphers
>
> "EECDH+ECDSA+AESGCM:EECDH+aRSA+AESGCM:EECDH+ECDSA+SHA384:EECDH+ECDSA+SHA256:EECDH+aRSA+SHA384:EECDH+aRSA+SHA256:EECDH:EDH+aRSA:-RC4:EECDH+aRSA+RC4:EECDH+RC4:EDH+aRSA+RC4:!aNULL:!eNULL:!LOW:!3DES:!MD5:!EXP:!PSK:!SRP:!DSS:RC4+SHA"
>     .....
> 8<----------------
>
> If you need Java 6 support, you need to do DH 1024 bit instead of 2048
> bit. (Pound's default for strong ciphers)
> If you're not interested in RC4 compatibility for older browsers, use a
> Ciphers line more like this:
>         Ciphers
>
> "EECDH+ECDSA+AESGCM:EECDH+aRSA+AESGCM:EECDH+ECDSA+SHA384:EECDH+ECDSA+SHA256:EECDH+aRSA+SHA384:EECDH+aRSA+SHA256:EECDH:EDH+aRSA:!RC4:!aNULL:!eNULL:!LOW:!3DES:!MD5:!EXP:!PSK:!SRP:!DSS"
>
> You need to be running OpenSSL 1.0+ for this to make any difference.
> Otherwise you won't have ECDHE or half the ciphers I've listed.  And for
> FALLBACK_SCSV reasons, you're going to want OpenSSL 1.0.1j or better, or
> if you're using 1.0.0 you'll want 1.0.0o or better.  (See US-CERT
> Advisory here
>
> https://www.us-cert.gov/ncas/current-activity/2014/10/16/OpenSSL-Patches-Four-Vulnerabilities
> )
>
> As neither Robert nor I supply RPMs or DEBs you'll have to check with
> your distribution's package maintainers to determine if and when those
> packages will be updated, whether they'll be running 2.7d, 2.6, or
> pcidss/v2.6. I know Martin Meredith (debian's maintainer) in the past
> has done similar but different patches to pound, so even though it says
> Pound 2.6, it may have additional vendor patches.  We have no way of
> knowing that.
>
> The only way to be sure about what version you're running and what
> features it has is to 1) compile it yourself, and 2) check the man page.
>
> Joe
>
>
>
>
>
>
>
>
> --
> To unsubscribe send an email with subject unsubscribe to [email protected]
> Please contact [email protected] for questions.
>



-- 
David Martineau
CTO
ContractPal, Inc.
801.494.1861 x120
david.martineau-A5OpC/[email protected] <e.david.martineau-A5OpC/[email protected]>

Index: src/http.c
IDEA additional info:
Subsystem: com.intellij.openapi.diff.impl.patch.CharsetEP
<+>UTF-8
===================================================================
--- src/http.c	(revision 3122)
+++ src/http.c	(revision )
@@ -533,7 +533,7 @@
 void
 do_http(thr_arg *arg)
 {
-    int                 cl_11, be_11, res, chunked, n, sock, no_cont, skip, conn_closed, force_10, sock_proto, is_rpc;
+    int                 cl_count,cl_11, be_11, res, chunked, n, sock, no_cont, skip, conn_closed, force_10, sock_proto, is_rpc;
     LISTENER            *lstn;
     SERVICE             *svc;
     BACKEND             *backend, *cur_backend, *old_backend;
@@ -708,7 +708,7 @@
         }
 
         /* check other headers */
-        for(chunked = 0, cont = L_1, n = 1; n < MAXHEADERS && headers[n]; n++) {
+        for(cl_count=0,chunked = 0, cont = L_1, n = 1; n < MAXHEADERS && headers[n]; n++) {
             /* no overflow - see check_header for details */
             switch(check_header(headers[n], buf)) {
             case HEADER_HOST:
@@ -734,11 +734,28 @@
                         chunked = 1;
                 break;
             case HEADER_CONTENT_LENGTH:
+                cl_count++;
+                if (cl_count>1)
+                {
+                    logmsg(LOG_WARNING, "(%lx) e501 bad multi-content-length request \"%s\" from %s", pthread_self(), request, caddr);
+                    err_reply(cl, h501, lstn->err501);
+                    free_headers(headers);
+                    clean_all();
+                    return;
+                }
+
                 if(chunked || cont >= 0L)
+                {
                     headers_ok[n] = 0;
-                else
+                }
+                else {
                     if((cont = ATOL(buf)) < 0L)
+                     {
                         headers_ok[n] = 0;
+                     }
+                    if(is_rpc == 1 && (cont < 0x20000L || cont > 0x80000000L))
+                        is_rpc = -1;
+                }
                 break;
             case HEADER_ILLEGAL:
                 if(lstn->log_level > 0) {
@@ -1401,8 +1418,12 @@
                 case HEADER_CONTENT_LENGTH:
                     cont = ATOL(buf);
                     /* treat RPC_OUT_DATA like reply without content-length */
-                    if(is_rpc == 0 && cont == 0x40000000L)
+                    if(is_rpc == 0) {
+                        if(cont >= 0x20000L && cont <= 0x80000000L)
                         cont = -1L;
+                        else
+                            is_rpc = -1;
+                    }
                     break;
                 case HEADER_LOCATION:
                     if(v_host[0] && need_rewrite(lstn->rewr_loc, buf, loc_path, v_host, lstn, cur_backend)) {