[Buildroot] [PATCH v2 1/1] paxtest: new package

Fri Jul 7 02:32:29 UTC 2017

Arnout,

On Thu, Jul 6, 2017 at 6:32 PM, Arnout Vandecappelle <arnout at mind.be> wrote:
>
>
> On 06-07-17 04:35, Matt Weber wrote:
> [snip]
>> diff --git a/package/paxtest/0002-paxtest-page-alignment-ARM-and-NIOS2-arch.patch b/package/paxtest/0002-paxtest-page-alignment-ARM-and-NIOS2-arch.patch
>> new file mode 100644
>> index 0000000..8e836d6
>> --- /dev/null
>> +++ b/package/paxtest/0002-paxtest-page-alignment-ARM-and-NIOS2-arch.patch
>> @@ -0,0 +1,49 @@
>> +From 70406ad5668a15fedce2ae1ed3bc4fad04d9f040 Mon Sep 17 00:00:00 2001
>> +From: Matt Weber <matthew.weber at rockwellcollins.com>
>> +Date: Wed, 5 Jul 2017 20:47:42 -0500
>> +Subject: [PATCH] paxtest: page alignment ARM and NIOS2 arch
>> +
>> +- Extended ARM range from ARMv6-v7 to also include v4-v7
>> +- Added NIOS2 arch to conditionally have smaller alignment
>> +
>> +Submitted Upstream to pageexec at freemail.hu.  Also posted a
>> +bug to both (Hardened) Suse and Gentoo's bugtrackers.
>> +https://bugzilla.opensuse.org/show_bug.cgi?id=1047422
>> +https://bugs.gentoo.org/show_bug.cgi?id=623946
>> +
>> +Signed-off-by: Matthew Weber <matthew.weber at rockwellcollins.com>
>> +---
>> + paxtest.h | 12 ++++++++++--
>> + 1 file changed, 10 insertions(+), 2 deletions(-)
>> +
>> +diff --git a/paxtest.h b/paxtest.h
>> +index 8623bfb..a230c1a 100644
>> +--- a/paxtest.h
>> ++++ b/paxtest.h
>> +@@ -5,13 +5,21 @@
>> + #include <unistd.h>
>> +
>> + /*
>> +- * ARMv6 and ARMv7 do not like 64k alignment, 32k is ok
>> ++ * ARMv4 through ARMv7 do not like 64k alignment, 32k is ok
>> +  */
>> +-#if defined(__arm__) && __ARM_ARCH >= 6 && __ARM_ARCH <= 7
>> ++#if defined(__arm__) && __ARM_ARCH >= 4 && __ARM_ARCH <= 7
>
>  >= 4 is always true, isn't it? I mean, ARMv3 has been dead for several decades, no?

Fair, that is true.  I can drop the >= 4.

>
>> + #define PAGE_SIZE_MAX  (32768)
>> + #else
>> + #define PAGE_SIZE_MAX       0x10000 /* 64k should cover most arches */
>> + #endif
>> ++
>> ++/*
>> ++ * NIOS2's assemblier doesn't like 64k alignment
>> ++ */
>> ++#if defined(__nios2_arch__)
>> ++#define PAGE_SIZE_MAX  (32768)
>> ++#endif
>> ++
>> + #ifndef __aligned
>> + #define __aligned(x)        __attribute__((aligned(x)))
>> + #endif
>> +--
>> +1.9.1
>> +
>> diff --git a/package/paxtest/Config.in b/package/paxtest/Config.in
>> new file mode 100644
>> index 0000000..5782d46
>> --- /dev/null
>> +++ b/package/paxtest/Config.in
>> @@ -0,0 +1,8 @@
>> +config BR2_PACKAGE_PAXTEST
>> +     bool "paxtest"
>> +     # No UCLIBC or MUSL because __NO_A_OUT_SUPPORT
>> +     depends on BR2_TOOLCHAIN_USES_GLIBC
>> +     help
>> +       PaX regression test suite
>> +
>> +       http://pax.grsecurity.net/docs
>
>  There should be a comment about the glibc requirement.

Opps, yes.  I'll add

comment "paxtest needs a glibc toolchain"
    depends on !BR2_TOOLCHAIN_USES_GLIBC

>
>> diff --git a/package/paxtest/paxtest.hash b/package/paxtest/paxtest.hash
>> new file mode 100644
>> index 0000000..c10566c
>> --- /dev/null
>> +++ b/package/paxtest/paxtest.hash
>> @@ -0,0 +1,2 @@
>> +# Locally computed:
>> +sha256       d553848431fd8c2ab6c8361b62e5cedfed1cc1d60088241f4a33d2af15dd667f        paxtest-0.9.15.tar.gz
>> diff --git a/package/paxtest/paxtest.mk b/package/paxtest/paxtest.mk
>> new file mode 100644
>> index 0000000..fe88571
>> --- /dev/null
>> +++ b/package/paxtest/paxtest.mk
>> @@ -0,0 +1,26 @@
>> +################################################################################
>> +#
>> +# paxtest
>> +#
>> +################################################################################
>> +
>> +PAXTEST_VERSION = 0.9.15
>> +PAXTEST_SOURCE = paxtest-$(PAXTEST_VERSION).tar.gz
>
>  This is the default and can be removed.

Thanks, the previous site we used had the dash as a underscore, didn't
realize when I updated it is back to normal.  Will remove.

>
>> +PAXTEST_SITE = https://www.grsecurity.net/~spender
>> +PAXTEST_LICENSE = GPL-2.0+
>> +PAXTEST_LICENSE_FILES = README
>> +
>> +# Not using TARGET_CONFIGURE_OPTS as this package needs to set it's own
>> +# compile/link flags to do checks for things like if -fPIE is working, etc
>
>  We typically do export TARGET_CONFIGURE_OPTS in the environment, which means
> the Makefile settings take precedence. Doesn't this work for the paxtest Makefile?
>

Agree, will remove statement and add $(TARGET_CONFIGURE_OPTS) as I did
not observed any build failures from the change.

>> +
>> +define PAXTEST_BUILD_CMDS
>> +     $(TARGET_MAKE_ENV) $(MAKE) -C $(@D) CC=$(TARGET_CC) LD=$(TARGET_LD) linux
>
>  TARGET_CC may consist of multiple words, so you have to quote this.
>
>  Does this package really use LD? That's usually annoying because it breaks
> situations that require specific architecture options to build correctly. Can
> you try if it works with LD=$(TARGET_CC) ?
>

It works using TARGET_CC instead.  Will make that update.

>> +endef
>> +
>> +define PAXTEST_INSTALL_TARGET_CMDS
>> +     $(TARGET_MAKE_ENV) $(MAKE) -C $(@D) DESTDIR=$(TARGET_DIR) \
>
>  Please use the same options as for building, i.e. include CC= and LD=, just in
> case something still gets built at install time.

Sure.

So my .mk ends up looking like:

PAXTEST_VERSION = 0.9.15
PAXTEST_SITE = https://www.grsecurity.net/~spender
PAXTEST_LICENSE = GPL-2.0+
PAXTEST_LICENSE_FILES = README

define PAXTEST_BUILD_CMDS
        $(TARGET_MAKE_ENV) $(TARGET_CONFIGURE_OPTS) $(MAKE) -C $(@D) \
                CC="$(TARGET_CC)" LD="$(TARGET_CC)" linux
endef

define PAXTEST_INSTALL_TARGET_CMDS
        $(TARGET_MAKE_ENV) $(TARGET_CONFIGURE_OPTS) $(MAKE) -C $(@D) \
                CC="$(TARGET_CC)" LD="$(TARGET_CC)" \
                DESTDIR=$(TARGET_DIR) \
                BINDIR="usr/bin" \
                RUNDIR="usr/lib" -f Makefile.psm install
endef

$(eval $(generic-package))

Appreciate the review!

Matt