Skip to contentSkip to conversation

ulogd fix for kernel 64bits/userspace 32bits system

PATCH In review Eric Leblond 2 participants 8 messages Oct 12, 2006 – Oct 13, 2006

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ulogd fix for kernel 64bits/userspace 32bits system

Hello,

This patch fixes ulogd for system with kernel space 64 bits and userspace 32 bits.

Auto detection of system type is taken from iptables.

BR,

Signature · 1 line
Eric Leblond <[email protected]>
Makefile.in +33 -0
Index: Makefile.in
===================================================================
@@ -31,6 +31,39 @@
 
 #  Normally You should not need to change anything below
 
+# Sparc64 hack
+ifeq ($(shell uname -m),sparc64)
+	POINTERTEST:=1
+	32bituser := $(shell echo -e "\#include <stdio.h>\n\#if !defined(__sparcv9) && !defined(__arch64__) && !defined(_LP64)\nuserspace_is_32bit\n\#endif" | $(CC) $(CFLAGS) -E - | grep userspace_is_32bit)
+	ifdef 32bituser
+		# The kernel is 64-bit, even though userspace is 32.
+		CFLAGS+=-DIPT_MIN_ALIGN=8 -DKERNEL_64_USERSPACE_32
+	else
+		EXT_LDFLAGS=-m elf64_sparc
+	endif
+endif
+
+# Alpha only has 64bit userspace and fails the test below
+ifeq ($(shell uname -m), alpha)
+	POINTERTEST:=1
+endif
+
+# Generic test if arch wasn't found above
+ifneq ($(POINTERTEST),1)
+	# Try to determine if kernel is 64bit and we are compiling for 32bit
+	ifeq ($(shell [ -a $(KERNEL_DIR)/include/asm ] && echo YES), YES)
+		64bitkernel := $(shell echo -e "\#include <asm/types.h>\n\#if BITS_PER_LONG == 64\nkernel_is_64bits\n\#endif" | $(CC) $(CFLAGS) -D__KERNEL__ -E - | grep kernel_is_64bits)
+		ifdef 64bitkernel
+			32bituser := $(shell echo -e "\#include <stdio.h>\n\#if !defined(__arch64__) && !defined(_LP64)\nuserspace_is_32bit\n\#endif" | $(CC) $(CFLAGS) -E - | grep userspace_is_32bit)
+			ifdef 32bituser
+				CFLAGS+=-DIPT_MIN_ALIGN=8 -DKERNEL_64_USERSPACE_32
+			endif
+		endif
+	else
+		CFLAGS+=-D_UNKNOWN_KERNEL_POINTER_SIZE
+	endif
+endif
+
 all: recurse ulogd
 
 .PHONY: distclean
include/linux/netfilter_ipv4/ipt_ULOG.h +62 -0

62 lines not on this page · raw message

include/libipulog/libipulog.h +2 -2
Index: libipulog/include/libipulog/libipulog.h
===================================================================
@@ -1,7 +1,7 @@
 #ifndef _LIBIPULOG_H
 #define _LIBIPULOG_H
 
-/* $Id: libipulog.h,v 1.6 2002/07/30 07:23:36 laforge Exp $ */
+/* $Id$ */
 
 #include <errno.h>
 #include <unistd.h>
@@ -12,7 +12,7 @@
 #include <asm/types.h>
 #include <linux/netlink.h>
 #include <net/if.h>
-#include <linux/netfilter_ipv4/ipt_ULOG.h>
+#include "linux/netfilter_ipv4/ipt_ULOG.h"
 
 /* FIXME: glibc sucks */
 #ifndef MSG_TRUNC 
Rules.make.in +1 -1
Index: Rules.make.in
===================================================================
@@ -21,7 +21,7 @@
 CFLAGS=@CFLAGS@ @CPPFLAGS@ -Wall
 CFLAGS+=-DULOGD_CONFIGFILE=\"$(ULOGD_CONFIGFILE)\"
 # doesn't work for subdirs
-#CFLAGS+=$(INCIPULOG) $(INCCONFFILE)
+CFLAGS+=$(INCIPULOG)
 CFLAGS+=-I/lib/modules/`uname -r`/build/include
 #CFLAGS+=@DEFS@
 #CFLAGS+=-g -DDEBUG -DDEBUG_MYSQL -DDEBUG_PGSQL
#1 lore raw
Quoted 6 lines · Eric Leblond

Hello,

This patch fixes ulogd for system with kernel space 64 bits and userspace 32 bits.

Auto detection of system type is taken from iptables.

We're actually going to remove that and I am very reluctant to add this.

- its a compile-time solution and therefor unusable for distributors

- it might get accidentally enabled by a distributor should he be crazy enough to compile packages in a mixed environment

- it will break if we ever get compat support in the kernel (which is admittedly quite unlikely)

So I really need some convincing :)

#2 lore raw
Quoted 9 lines · Patrick McHardy
Quoted 1 line · Eric Leblond

Hello,

We're actually going to remove that and I am very reluctant to add this.

- its a compile-time solution and therefor unusable for distributors

- it might get accidentally enabled by a distributor should he be crazy enough to compile packages in a mixed environment

Both points seem linked with autodetection of system. I agree that it is potentially dangerous.

In fact, it may be better to use a configure flag approach ? What do you think ?

BR,

Signature · 3 lines
Éric Leblond, [email protected]
Téléphone : 01 44 89 46 39, Fax : 01 44 89 45 01
INL, http://www.inl.fr
#3 lore raw
Quoted 8 lines · Eric Leblond
Quoted 4 lines · Patrick McHardy

- its a compile-time solution and therefor unusable for distributors

- it might get accidentally enabled by a distributor should he be crazy enough to compile packages in a mixed environment

Quoted 5 lines

Both points seem linked with autodetection of system. I agree that it is potentially dangerous.

In fact, it may be better to use a configure flag approach ? What do you think ?

That sounds better - ideally with a big warning that states that it is just a workaround and _will_ break in case real compatibility is introduced to the kernel.

#4 lore raw
Quoted 14 lines · Patrick McHardy
Quoted 12 lines · Eric Leblond
Quoted 4 lines · Patrick McHardy

- its a compile-time solution and therefor unusable for distributors

- it might get accidentally enabled by a distributor should he be crazy enough to compile packages in a mixed environment

Both points seem linked with autodetection of system. I agree that it is potentially dangerous.

Quoted 7 lines
Quoted 2 lines

In fact, it may be better to use a configure flag approach ? What do you think ?

That sounds better - ideally with a big warning that states that it is just a workaround and _will_ break in case real compatibility is introduced to the kernel.

BTW, maybe its even possible to do runtime detection? That would also fix the first issue.

#5 lore raw

Resend: Patch : ulogd fix for kernel 64bits/userspace 32bits system

In reply to Patrick McHardy
Quoted 8 lines · Patrick McHardy
Quoted 2 lines · Eric Leblond

In fact, it may be better to use a configure flag approach ? What do you think ?

That sounds better - ideally with a big warning that states that it is just a workaround and _will_ break in case real compatibility is introduced to the kernel.

This is done.

You can now do :

	./configure --with-kernel-64-user-32

to activate the workaround.

It displays :

configure: WARNING: The use of the flag kernel-64-user-32 could interfere with kernel evolution. Use it at your own risk.

BR,

Signature · 1 line
Eric Leblond <[email protected]>
configure.in +16 -0
Index: configure.in
===================================================================
@@ -31,6 +31,8 @@
 
 DATABASE_DRIVERS=""
 
+KERNEL64_USERSPACE32=""
+
 dnl
 dnl test for MySQL
 dnl
@@ -250,7 +252,19 @@
    AC_MSG_WARN(the use of --with-sqlite3-log-ip-as-string is discouraged)
 ])
 
+dnl
+dnl Kernel 64
+dnl
 
+AC_ARG_WITH(kernel-64-user-32,
+ --with-kernel-64-user-32		Use this flag to compile on system where kernel is 64 bits
+					userspace is 32.
+,[
+   KERNEL64_USERSPACE32="-DKERNEL_64_USERSPACE_32"
+   AC_MSG_WARN(The use of the flag kernel-64-user-32 could interfere with kernel evolution. Use it at your own risk.)
+])
+
+
 AC_SUBST(DATABASE_DIR)
 AC_SUBST(DATABASE_LIB)
 AC_SUBST(DATABASE_LIB_DIR)
@@ -262,6 +276,8 @@
 AC_SUBST(DATABASE_DRIVERS)
 AC_SUBST(HAVE_PCAP_H)
 
+AC_SUBST(KERNEL64_USERSPACE32)
+
 AM_CONDITIONAL(HAVE_MYSQL, test x$mysqldir != x)
 AM_CONDITIONAL(HAVE_PGSQL, test x$pgsqldir != x)
 AM_CONDITIONAL(HAVE_SQLITE3, test x$sqlite3dir != x)
include/linux/netfilter_ipv4/ipt_ULOG.h +62 -0

62 lines not on this page · raw message

include/libipulog/libipulog.h +2 -2
Index: libipulog/include/libipulog/libipulog.h
===================================================================
@@ -1,7 +1,7 @@
 #ifndef _LIBIPULOG_H
 #define _LIBIPULOG_H
 
-/* $Id: libipulog.h,v 1.6 2002/07/30 07:23:36 laforge Exp $ */
+/* $Id$ */
 
 #include <errno.h>
 #include <unistd.h>
@@ -12,7 +12,7 @@
 #include <asm/types.h>
 #include <linux/netlink.h>
 #include <net/if.h>
-#include <linux/netfilter_ipv4/ipt_ULOG.h>
+#include "linux/netfilter_ipv4/ipt_ULOG.h"
 
 /* FIXME: glibc sucks */
 #ifndef MSG_TRUNC 
Rules.make.in +2 -1
Index: Rules.make.in
===================================================================
@@ -20,8 +20,9 @@
 
 CFLAGS=@CFLAGS@ @CPPFLAGS@ -Wall
 CFLAGS+=-DULOGD_CONFIGFILE=\"$(ULOGD_CONFIGFILE)\"
+CFLAGS+=@KERNEL64_USERSPACE32@
 # doesn't work for subdirs
-#CFLAGS+=$(INCIPULOG) $(INCCONFFILE)
+CFLAGS+=$(INCIPULOG)
 CFLAGS+=-I/lib/modules/`uname -r`/build/include
 #CFLAGS+=@DEFS@
 #CFLAGS+=-g -DDEBUG -DDEBUG_MYSQL -DDEBUG_PGSQL
#6 lore raw

Re: Resend: Patch : ulogd fix for kernel 64bits/userspace 32bits system

Quoted 7 lines · Eric Leblond

You can now do :

	./configure --with-kernel-64-user-32

to activate the workaround.

It displays :

configure: WARNING: The use of the flag kernel-64-user-32 could interfere with kernel evolution. Use it at your own risk.

The patch looks good, but doesn't apply:

(Stripping trailing CRs from patch.)
patching file configure.in
(Stripping trailing CRs from patch.)
patching file libipulog/include/linux/netfilter_ipv4/ipt_ULOG.h (Stripping trailing CRs from patch.)
patching file libipulog/include/libipulog/libipulog.h Hunk #1 FAILED at 1.
1 out of 2 hunks FAILED -- saving rejects to file libipulog/include/libipulog/libipulog.h.rej (Stripping trailing CRs from patch.)
patching file Rules.make.in

Please rediff against current SVN.

#7 lore raw

Re: Resend: Patch : ulogd fix for kernel 64bits/userspace 32bits system

Quoted 14 lines · Patrick McHardy

The patch looks good, but doesn't apply:

(Stripping trailing CRs from patch.)
patching file configure.in
(Stripping trailing CRs from patch.)
patching file libipulog/include/linux/netfilter_ipv4/ipt_ULOG.h (Stripping trailing CRs from patch.)
patching file libipulog/include/libipulog/libipulog.h Hunk #1 FAILED at 1.
1 out of 2 hunks FAILED -- saving rejects to file libipulog/include/libipulog/libipulog.h.rej (Stripping trailing CRs from patch.)
patching file Rules.make.in

Quoted 1 line

Please rediff against current SVN.

Never mind, it was this part:

-/* $Id: libipulog.h,v 1.6 2002/07/30 07:23:36 laforge Exp $ */ +/* $Id$ */

I fixed it up manually. Thanks Eric.