Skip to contentSkip to conversation

Allow building libtirpc directly against GSSAPI

PATCH Simo Sorce 5 participants 10 messages Mar 26, 2013 – Apr 2, 2013

Subsystems (2)
lore
Share & tools
lore

Simo Sorce

author
#0 lore raw

Libgssglue is not really useful anymore, it is a sort of middleman that wraps the actual GSSAPI that is already pluggable/extensible via shared modules.

In particular libgssglue interferes with the workings of gss-proxy in my case.

The attached patch makes building against libgssglue optional and defaults to not build against libgssglue and instead builds directly against the native GSSAPI.

./configure --enable-gss
will now build against GSSAPI

./configure --enable-gss --with-gssglue
will keep building against libgssglue in case someone still needs it for whatever reason.

Simo.

Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
Quoted 1 line

From 8518f6857b928e9ec615f9ca5f79f98d13db4f6d Mon Sep 17 00:00:00 2001

From: Simo Sorce <[email protected]>
Date: Tue, 26 Mar 2013 10:15:56 -0400
Subject: [PATCH] Switch to use standard GSSAPI by default

Make libgssglue configurable still but disabled by default. There is no reason to use libgssglue anymore, and modern gssapi supports all needed features for libtirpc and its dependencies.

---
 configure.ac    | 23 +++++++++++++++++++----
 src/Makefile.am |  4 ++--
 2 files changed, 21 insertions(+), 6 deletions(-)
configure.ac +19 -4
index 40dce96..4a4adba 100644
@@ -5,15 +5,30 @@ AC_CONFIG_SRCDIR([src/auth_des.c])
 AC_CONFIG_MACRO_DIR([m4])
 
 AC_ARG_ENABLE(gss,[  --enable-gss            Turn on gss api], [case "${enableval}" in
-        yes) gss=true ; AC_CHECK_LIB([gssapi],[gss_init_sec_context]) ;;
+        yes) gss=true ;;
         no)  gss=false ;;
         *) AC_MSG_ERROR(bad value ${enableval} for --enable-gss) ;;
       esac],[gss=false])
 AM_CONDITIONAL(GSS, test x$gss = xtrue)
+AC_ARG_WITH(gssglue,
+		[  --with-gssglue        Use libgssglue],
+		 [case "${enableval}" in
+		  yes) gssglue=true ;;
+		  no)  gssglue=false ;;
+		  *) AC_MSG_ERROR(bad value ${enableval} for --with-gssglue) ;;
+		  esac],
+		[gssglue=false])
+AM_CONDITIONAL(USEGSSGLUE, test x$gssglue = xtrue)
 if test x$gss = xtrue; then
-	AC_DEFINE(HAVE_LIBGSSAPI, 1, [])
-	PKG_CHECK_MODULES(GSSGLUE, libgssglue, [],
-	AC_MSG_ERROR([Unable to locate information required to use libgssglue.]))
+	if test x$gssglue = xtrue; then
+		PKG_CHECK_MODULES(GSSAPI, libgssglue, [],
+		AC_MSG_ERROR([Unable to locate information required to use libgssglue.]))
+	else
+		GSSAPI_CFLAGS=`krb5-config --cflags gssapi`
+		GSSAPI_LIBS=`krb5-config --libs gssapi`
+		AC_SUBST([GSSAPI_CFLAGS])
+		AC_SUBST([GSSAPI_LIBS])
+	fi
 fi
 AC_ARG_ENABLE(ipv6,
 	[AC_HELP_STRING([--disable-ipv6], [Disable IPv6 support @<:@default=no@:>@])],
src/Makefile.am +2 -2
index 66350f5..2dd7768 100644
@@ -58,8 +58,8 @@ libtirpc_la_SOURCES += xdr.c xdr_rec.c xdr_array.c xdr_float.c xdr_mem.c xdr_ref
 ## Secure-RPC
 if GSS
     libtirpc_la_SOURCES += auth_gss.c authgss_prot.c svc_auth_gss.c
-    libtirpc_la_LDFLAGS += $(GSSGLUE_LIBS)
-    libtirpc_la_CFLAGS = -DHAVE_RPCSEC_GSS $(GSSGLUE_CFLAGS)
+    libtirpc_la_LDFLAGS += $(GSSAPI_LIBS)
+    libtirpc_la_CFLAGS = -DHAVE_RPCSEC_GSS $(GSSAPI_CFLAGS)
 endif
 
 ## libtirpc_a_SOURCES += key_call.c key_prot_xdr.c getpublickey.c
Signature · 1 line
1.8.1.4
#1 lore raw
Quoted 11 lines · Simo Sorce

Libgssglue is not really useful anymore, it is a sort of middleman that wraps the actual GSSAPI that is already pluggable/extensible via shared modules.

In particular libgssglue interferes with the workings of gss-proxy in my case.

The attached patch makes building against libgssglue optional and defaults to not build against libgssglue and instead builds directly against the native GSSAPI.

Quoted 9 lines

./configure --enable-gss
will now build against GSSAPI

./configure --enable-gss --with-gssglue
will keep building against libgssglue in case someone still needs it for whatever reason.

Simo.

Committed....

steved.

#2 lore raw
In reply to Simo Sorce
Quoted 11 lines · Simo Sorce

Libgssglue is not really useful anymore, it is a sort of middleman that wraps the actual GSSAPI that is already pluggable/extensible via shared modules.

In particular libgssglue interferes with the workings of gss-proxy in my case.

The attached patch makes building against libgssglue optional and defaults to not build against libgssglue and instead builds directly against the native GSSAPI.

Quoted 9 lines

./configure --enable-gss
will now build against GSSAPI

./configure --enable-gss --with-gssglue
will keep building against libgssglue in case someone still needs it for whatever reason.

Simo.

Won't that be a backward compatibility issue?

Signature · 6 lines
Trond Myklebust
Linux NFS client maintainer

NetApp
[email protected]
www.netapp.com
author
#3 lore raw
Quoted 15 lines · Trond
Quoted 14 lines · Simo Sorce

Libgssglue is not really useful anymore, it is a sort of middleman that wraps the actual GSSAPI that is already pluggable/extensible via shared modules.

In particular libgssglue interferes with the workings of gss-proxy in my case.

The attached patch makes building against libgssglue optional and defaults to not build against libgssglue and instead builds directly against the native GSSAPI.

./configure --enable-gss
will now build against GSSAPI

Quoted 8 lines
Quoted 6 lines

./configure --enable-gss --with-gssglue
will keep building against libgssglue in case someone still needs it for whatever reason.

Simo.

Won't that be a backward compatibility issue?

What are you worried about exactly ? Is there a use case we should know about ?

Simo.

Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
#4 lore raw
Quoted 20 lines · Simo Sorce
Quoted 19 lines · Trond
Quoted 18 lines · Simo Sorce

Libgssglue is not really useful anymore, it is a sort of middleman that wraps the actual GSSAPI that is already pluggable/extensible via shared modules.

In particular libgssglue interferes with the workings of gss-proxy in my case.

The attached patch makes building against libgssglue optional and defaults to not build against libgssglue and instead builds directly against the native GSSAPI.

./configure --enable-gss
will now build against GSSAPI

./configure --enable-gss --with-gssglue
will keep building against libgssglue in case someone still needs it for whatever reason.

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

Simo.

Won't that be a backward compatibility issue?

What are you worried about exactly ? Is there a use case we should know about ?

Building nfs-utils against existing older setups. There are a lot of users out there of 4-5 year old distros (including ones maintained by your employer).

Why isn't it safe to assume that if someone has libgssglue installed, then we should be building nfs-utils against it?

Signature · 6 lines
Trond Myklebust
Linux NFS client maintainer

NetApp
[email protected]
www.netapp.com
author
#5 lore raw
Quoted 29 lines · Trond
Quoted 20 lines · Simo Sorce
Quoted 19 lines · Trond
Quoted 18 lines · Simo Sorce

Libgssglue is not really useful anymore, it is a sort of middleman that wraps the actual GSSAPI that is already pluggable/extensible via shared modules.

In particular libgssglue interferes with the workings of gss-proxy in my case.

The attached patch makes building against libgssglue optional and defaults to not build against libgssglue and instead builds directly against the native GSSAPI.

./configure --enable-gss
will now build against GSSAPI

./configure --enable-gss --with-gssglue
will keep building against libgssglue in case someone still needs it for whatever reason.

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

Simo.

Won't that be a backward compatibility issue?

What are you worried about exactly ? Is there a use case we should know about ?

Quoted 6 lines

Building nfs-utils against existing older setups. There are a lot of users out there of 4-5 year old distros (including ones maintained by your employer).

Why isn't it safe to assume that if someone has libgssglue installed, then we should be building nfs-utils against it?

A foillowing patch for nfs-utils is coming, both the libtirpc and nfs-utils patches will make building against libgssglue optional.

I see no value for selecting libgssglue by default though. libgssglue does nothing but load the underlying gssapi library by default anyway and adds nothing to that.

I am leaving the option to compile against libgssglue exactly for the odd case where someone has a good reason to use it (very, very old systems). In that case all they need to do is to run configure with --enable-gss --with-gssglue

Simo.

Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
#6 lore raw

Doesn't nfs-utils build without gssglue if libgssglue is not installed? If that's true I see no reason to change the default. If you don't want gssglue, uninstall it. If you have it installed and want to build without it, use --without-gssglue.

author
#7 lore raw
Quoted 1 line · Jim Rees

Doesn't nfs-utils build without gssglue if libgssglue is not installed?

Not really, both libtirpc needs --enable-gss, if yo pass that they check only for libgssglue right now and fail if not available.

Quoted 3 lines

If that's true I see no reason to change the default. If you don't want gssglue, uninstall it. If you have it installed and want to build without it, use --without-gssglue.

I am changing the default to build against system GSSAPI libraries when --enable-gss is passed in, if you want to force to build against libgssglue for some specific reason you use --with-gssglue

I believe building against system libraries instead of a shim is the better default in the general case.

Simo.

Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
#8 lore raw
In reply to Simo Sorce

Simo Sorce <simo@...> writes:

Hi,

If you've already mentioned the gssglue issue, there's a related one, namely, building nfs-utils against Heimdal.

Currently, the out of the box Heimdal support is broken, and most of the breakage comes out of the gssglue.

I'm looking at fixing nfs-utils to support Heimdal properly - currently my only remaining problem is to fix the configure and pkg-config scripts in both nfs- utils and libgssglue (if this one is not dropped for good, and I personally think it should be; small, icky library on no real use).

Here is the code patch I'm using for my Heimdal build:

diff -ur nfs-utils-1.2.6.orig/utils/gssd/context_lucid.c nfs-utils- 1.2.6/utils/gssd/context_lucid.c
--- nfs-utils-1.2.6.orig/utils/gssd/context_lucid.c 2012-05-15
00:40:52.000000000 +1000
+++ nfs-utils-1.2.6/utils/gssd/context_lucid.c 2013-03-26 19:03:10.096586556
+1100
@@ -266,10 +266,10 @@
int retcode = 0;

printerr(2, "DEBUG: %s: lucid version!\n", __FUNCTION__);
- maj_stat = gss_export_lucid_sec_context(&min_stat, &ctx,
- 1, &return_ctx);
+ maj_stat = gss_krb5_export_lucid_sec_context(&min_stat, &ctx,
+ - 1, &return_ctx);
if (maj_stat != GSS_S_COMPLETE) {
- pgsserr("gss_export_lucid_sec_context",
+ pgsserr("gss_krb5_export_lucid_sec_context",
maj_stat, min_stat, &krb5oid);
goto out_err;
}
@@ -302,9 +302,9 @@
else
retcode = prepare_krb5_rfc4121_buffer(lctx, buf, endtime);

- maj_stat = gss_free_lucid_sec_context(&min_stat, ctx, return_ctx);
+ maj_stat = gss_krb5_free_lucid_sec_context(&min_stat, ctx);
if (maj_stat != GSS_S_COMPLETE) {
- pgsserr("gss_free_lucid_sec_context",
+ pgsserr("gss_krb5_free_lucid_sec_context",
maj_stat, min_stat, &krb5oid);
printerr(0, "WARN: failed to free lucid sec context\n");
}
diff -ur nfs-utils-1.2.6.orig/utils/gssd/krb5_util.c nfs-utils- 1.2.6/utils/gssd/krb5_util.c
--- nfs-utils-1.2.6.orig/utils/gssd/krb5_util.c 2012-05-15 00:40:52.000000000
+1000
+++ nfs-utils-1.2.6/utils/gssd/krb5_util.c 2013-03-26 19:18:40.204045067
+1100
@@ -115,7 +115,7 @@
#include <errno.h>
#include <time.h>
#include <gssapi/gssapi.h>
-#ifdef USE_PRIVATE_KRB5_FUNCTIONS
+#if defined(USE_PRIVATE_KRB5_FUNCTIONS) || defined(HAVE_HEIMDAL)
#include <gssapi/gssapi_krb5.h>
#endif
#include <krb5.h>
@@ -936,9 +936,38 @@
{
krb5_error_code ret;
krb5_creds creds;
- krb5_cc_cursor cur;
int found = 0;

+#if defined (HAVE_HEIMDAL)
+ krb5_creds pattern;
+ krb5_const_realm client_realm;
+
+ krb5_cc_clear_mcred(&pattern);
+
+ client_realm = krb5_principal_get_realm(context, principal);
+
+ ret = krb5_make_principal(context, &pattern.server,
+ client_realm, KRB5_TGS_NAME, client_realm,
+ NULL);
+ if (ret)
+ krb5_err(context, 1, ret, "krb5_make_principal");
+ pattern.client = principal;
+
+ ret = krb5_cc_retrieve_cred(context, ccache, 0, &pattern, &creds);
+ krb5_free_principal(context, pattern.server);
+ if (ret) {
+ if (ret == KRB5_CC_END)
+ return 1;
+ krb5_err(context, 1, ret, "krb5_cc_retrieve_cred");
+ }
+
+ found = creds.times.endtime > time(NULL);
+
+ krb5_free_cred_contents (context, &creds);
+#else
+ krb5_cc_cursor cur;
+
+
ret = krb5_cc_start_seq_get(context, ccache, &cur); if (ret)
return 0;
@@ -958,7 +987,7 @@
krb5_free_cred_contents(context, &creds);
}
krb5_cc_end_seq_get(context, ccache, &cur);
-
+#endif
return found;
}

@@ -1278,7 +1307,7 @@
return strdup(error_message(code));
#else
if (context != NULL)
- return strdup(krb5_get_err_text(context, code));
+ return strdup(krb5_get_error_message(context, code));
else
return strdup(error_message(code));
#endif
@@ -1347,11 +1376,11 @@
* list of supported enctypes, use local default here. */
if (krb5_enctypes == NULL || limit_to_legacy_enctypes)
- maj_stat = gss_set_allowable_enctypes(&min_stat, credh,
- &krb5oid, num_enctypes, enctypes);
+ maj_stat = gss_krb5_set_allowable_enctypes(&min_stat, credh,
+ num_enctypes, enctypes);
else
- maj_stat = gss_set_allowable_enctypes(&min_stat, credh,
- &krb5oid, num_krb5_enctypes,
krb5_enctypes);
+ maj_stat = gss_krb5_set_allowable_enctypes(&min_stat, credh,
+ num_krb5_enctypes, krb5_enctypes);

if (maj_stat != GSS_S_COMPLETE) {
pgsserr("gss_set_allowable_enctypes",

#9 lore raw

CC-ing Simo since he is not on this list...

Quoted 91 lines · Alex Dubov

Simo Sorce <simo@...> writes:

Hi,

If you've already mentioned the gssglue issue, there's a related one, namely, building nfs-utils against Heimdal.

Currently, the out of the box Heimdal support is broken, and most of the breakage comes out of the gssglue.

I'm looking at fixing nfs-utils to support Heimdal properly - currently my only remaining problem is to fix the configure and pkg-config scripts in both nfs- utils and libgssglue (if this one is not dropped for good, and I personally think it should be; small, icky library on no real use).

Here is the code patch I'm using for my Heimdal build:

diff -ur nfs-utils-1.2.6.orig/utils/gssd/context_lucid.c nfs-utils- 1.2.6/utils/gssd/context_lucid.c
--- nfs-utils-1.2.6.orig/utils/gssd/context_lucid.c 2012-05-15
00:40:52.000000000 +1000
+++ nfs-utils-1.2.6/utils/gssd/context_lucid.c 2013-03-26 19:03:10.096586556
+1100

(quoted diff) +5 -5
@@ -266,10 +266,10 @@
 	int retcode = 0;
 
 	printerr(2, "DEBUG: %s: lucid version!\n", __FUNCTION__);
-	maj_stat = gss_export_lucid_sec_context(&min_stat, &ctx,
-						1, &return_ctx);
+	maj_stat = gss_krb5_export_lucid_sec_context(&min_stat, &ctx,
+	- 						1, &return_ctx);
 	if (maj_stat != GSS_S_COMPLETE) {
-		pgsserr("gss_export_lucid_sec_context",
+		pgsserr("gss_krb5_export_lucid_sec_context",
 			maj_stat, min_stat, &krb5oid);
 		goto out_err;
 	}
@@ -302,9 +302,9 @@
 	else
 		retcode = prepare_krb5_rfc4121_buffer(lctx, buf, endtime);
 
-	maj_stat = gss_free_lucid_sec_context(&min_stat, ctx, return_ctx);
+	maj_stat = gss_krb5_free_lucid_sec_context(&min_stat, ctx);
 	if (maj_stat != GSS_S_COMPLETE) {
-		pgsserr("gss_free_lucid_sec_context",
+		pgsserr("gss_krb5_free_lucid_sec_context",
 			maj_stat, min_stat, &krb5oid);
 		printerr(0, "WARN: failed to free lucid sec context\n");
 	}

diff -ur nfs-utils-1.2.6.orig/utils/gssd/krb5_util.c nfs-utils- 1.2.6/utils/gssd/krb5_util.c
--- nfs-utils-1.2.6.orig/utils/gssd/krb5_util.c 2012-05-15 00:40:52.000000000
+1000
+++ nfs-utils-1.2.6/utils/gssd/krb5_util.c 2013-03-26 19:18:40.204045067
+1100

(quoted diff) +21 -2
@@ -115,7 +115,7 @@
 #include <errno.h>
 #include <time.h>
 #include <gssapi/gssapi.h>
-#ifdef USE_PRIVATE_KRB5_FUNCTIONS
+#if defined(USE_PRIVATE_KRB5_FUNCTIONS) || defined(HAVE_HEIMDAL)
 #include <gssapi/gssapi_krb5.h>
 #endif
 #include <krb5.h>
@@ -936,9 +936,38 @@
 {
 	krb5_error_code ret;
 	krb5_creds creds;
-	krb5_cc_cursor cur;
 	int found = 0;
 
+#if defined (HAVE_HEIMDAL)
+	krb5_creds pattern;
+	krb5_const_realm client_realm;
+
+	krb5_cc_clear_mcred(&pattern);
+
+	client_realm = krb5_principal_get_realm(context, principal);
+
+	ret = krb5_make_principal(context, &pattern.server,
+				  client_realm, KRB5_TGS_NAME, client_realm,
+				  NULL);
+	if (ret)
+	  krb5_err(context, 1, ret, "krb5_make_principal");
+	pattern.client = principal;
+
+	ret = krb5_cc_retrieve_cred(context, ccache, 0, &pattern, &creds);
+	krb5_free_principal(context, pattern.server);
+	if (ret) {
+	  if (ret == KRB5_CC_END)
+            return 1;

Review line 970

Quoted 10 lines
(quoted diff) +10 -0
+	  krb5_err(context, 1, ret, "krb5_cc_retrieve_cred");
+	}
+
+	found = creds.times.endtime > time(NULL);
+
+	krb5_free_cred_contents (context, &creds);
+#else
+	krb5_cc_cursor cur;
+
+

This bug huge ifdef is ugly... ;-) Can we redefine what check_for_tgt() contains depending on HAVE_HEIMDAL and HAVE_KRB5?

Quoted 13 lines
(quoted diff) +1 -1
 	ret = krb5_cc_start_seq_get(context, ccache, &cur);
 	if (ret) 
 		return 0;
@@ -958,7 +987,7 @@
 		krb5_free_cred_contents(context, &creds);
 	}
 	krb5_cc_end_seq_get(context, ccache, &cur);
-
+#endif
 	return found;
 }
 

Review line 1310

Quoted 5 lines
(quoted diff) +1 -1
@@ -1278,7 +1307,7 @@
 	return strdup(error_message(code));
 #else
 	if (context != NULL)
-		return strdup(krb5_get_err_text(context, code));
+		return strdup(krb5_get_error_message(context, code));

Not sure why this is needed since they are both define in the krb5 libs Does krb5_get_error_message() give better error messages?

steved.

Quoted 26 lines
(quoted diff) +4 -5
 	else
 		return strdup(error_message(code));
 #endif
@@ -1347,11 +1376,11 @@
 	 * list of supported enctypes, use local default here.
 	 */
 	if (krb5_enctypes == NULL || limit_to_legacy_enctypes)
-		maj_stat = gss_set_allowable_enctypes(&min_stat, credh,
-					&krb5oid, num_enctypes, enctypes);
+		maj_stat = gss_krb5_set_allowable_enctypes(&min_stat, credh,
+					num_enctypes, enctypes);
 	else
-		maj_stat = gss_set_allowable_enctypes(&min_stat, credh,
-					&krb5oid, num_krb5_enctypes, 
 krb5_enctypes);
+		maj_stat = gss_krb5_set_allowable_enctypes(&min_stat, credh,
+					num_krb5_enctypes, krb5_enctypes);
 	if (maj_stat != GSS_S_COMPLETE) {
 		pgsserr("gss_set_allowable_enctypes",
 
 
--

To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html