[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAK7LNATDbxc+3HQ6zoSk9t-Lkf4MSNmEUN6S5EqoVWnBQw_K6g@mail.gmail.com>
Date: Sat, 3 May 2025 15:13:06 +0900
From: Masahiro Yamada <masahiroy@...nel.org>
To: Kees Cook <kees@...nel.org>
Cc: Nathan Chancellor <nathan@...nel.org>, Nicolas Schier <nicolas.schier@...ux.dev>,
Petr Pavlu <petr.pavlu@...e.com>, Sebastian Andrzej Siewior <bigeasy@...utronix.de>,
linux-kbuild@...r.kernel.org, Justin Stitt <justinstitt@...gle.com>,
Marco Elver <elver@...gle.com>, Andrey Konovalov <andreyknvl@...il.com>,
Andrey Ryabinin <ryabinin.a.a@...il.com>, Richard Weinberger <richard@....at>,
Anton Ivanov <anton.ivanov@...bridgegreys.com>, Johannes Berg <johannes@...solutions.net>,
linux-kernel@...r.kernel.org, linux-hardening@...r.kernel.org,
kasan-dev@...glegroups.com, linux-um@...ts.infradead.org
Subject: Re: [PATCH v2 2/3] randstruct: Force full rebuild when seed changes
On Sat, May 3, 2025 at 7:54 AM Kees Cook <kees@...nel.org> wrote:
>
> While the randstruct GCC plugin was being rebuilt if the randstruct
> seed changed, Clangs build did not notice the change. Include the hash
> header directly so that it becomes a universal build dependency and full
> rebuilds will happen if it changes.
>
> Since we cannot use "-include ..." as the randstruct flags are removed
> via "filter-out" (which would cause all instances of "-include" to be
> removed), use the existing -DRANDSTRUCT to control the header inclusion
> via include/linux/compiler-version.h. Universally add a -I for the
> scripts/basic directory, where header exists. The UM build requires that
> the -I be explicitly added.
>
> Signed-off-by: Kees Cook <kees@...nel.org>
> ---
> Cc: Masahiro Yamada <masahiroy@...nel.org>
> Cc: Nathan Chancellor <nathan@...nel.org>
> Cc: Nicolas Schier <nicolas.schier@...ux.dev>
> Cc: Petr Pavlu <petr.pavlu@...e.com>
> Cc: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
> Cc: <linux-kbuild@...r.kernel.org>
> ---
> Makefile | 1 +
> arch/um/Makefile | 1 +
> include/linux/compiler-version.h | 3 +++
> include/linux/vermagic.h | 1 -
> scripts/basic/Makefile | 11 ++++++-----
> 5 files changed, 11 insertions(+), 6 deletions(-)
>
> diff --git a/Makefile b/Makefile
> index 5aa9ee52a765..cef652227843 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -567,6 +567,7 @@ LINUXINCLUDE := \
> -I$(objtree)/arch/$(SRCARCH)/include/generated \
> -I$(srctree)/include \
> -I$(objtree)/include \
> + -I$(objtree)/scripts/basic \
Now you are adding random header search paths everywhere.
This is very hacky.
I recommend keeping <generated/randstruct_hash.h>
Then, -I$(objtree)/scripts/basic is unneeded.
> $(USERINCLUDE)
>
> KBUILD_AFLAGS := -D__ASSEMBLY__ -fno-PIE
> diff --git a/arch/um/Makefile b/arch/um/Makefile
> index 8cc0f22ebefa..38f6024e75d7 100644
> --- a/arch/um/Makefile
> +++ b/arch/um/Makefile
> @@ -73,6 +73,7 @@ USER_CFLAGS = $(patsubst $(KERNEL_DEFINES),,$(patsubst -I%,,$(KBUILD_CFLAGS))) \
> -D_FILE_OFFSET_BITS=64 -idirafter $(srctree)/include \
> -idirafter $(objtree)/include -D__KERNEL__ -D__UM_HOST__ \
> -I$(objtree)/scripts/gcc-plugins \
> + -I$(objtree)/scripts/basic \
> -include $(srctree)/include/linux/compiler-version.h \
> -include $(srctree)/include/linux/kconfig.h
>
> diff --git a/include/linux/compiler-version.h b/include/linux/compiler-version.h
> index 08943df04ebb..05d555320a0f 100644
> --- a/include/linux/compiler-version.h
> +++ b/include/linux/compiler-version.h
> @@ -16,3 +16,6 @@
> #ifdef GCC_PLUGINS_ENABLED
> #include "gcc-plugins-deps.h"
> #endif
> +#ifdef RANDSTRUCT
> +#include "randstruct_hash.h"
> +#endif
> diff --git a/include/linux/vermagic.h b/include/linux/vermagic.h
> index 939ceabcaf06..335c360d4f9b 100644
> --- a/include/linux/vermagic.h
> +++ b/include/linux/vermagic.h
> @@ -33,7 +33,6 @@
> #define MODULE_VERMAGIC_MODVERSIONS ""
> #endif
> #ifdef RANDSTRUCT
> -#include <generated/randstruct_hash.h>
> #define MODULE_RANDSTRUCT "RANDSTRUCT_" RANDSTRUCT_HASHED_SEED
> #else
> #define MODULE_RANDSTRUCT
> diff --git a/scripts/basic/Makefile b/scripts/basic/Makefile
> index dd289a6725ac..31637ce4dc5c 100644
> --- a/scripts/basic/Makefile
> +++ b/scripts/basic/Makefile
> @@ -8,9 +8,10 @@ hostprogs-always-y += fixdep
> # before running a Clang kernel build.
> gen-randstruct-seed := $(srctree)/scripts/gen-randstruct-seed.sh
> quiet_cmd_create_randstruct_seed = GENSEED $@
> -cmd_create_randstruct_seed = \
> - $(CONFIG_SHELL) $(gen-randstruct-seed) \
> - $@ $(objtree)/include/generated/randstruct_hash.h
> -$(obj)/randstruct.seed: $(gen-randstruct-seed) FORCE
> + cmd_create_randstruct_seed = $(CONFIG_SHELL) $(gen-randstruct-seed) \
> + $(obj)/randstruct.seed $(obj)/randstruct_hash.h
> +
> +$(obj)/randstruct_hash.h $(obj)/randstruct.seed: $(gen-randstruct-seed) FORCE
> $(call if_changed,create_randstruct_seed)
This is wrong.
$(obj)/randstruct_hash.h $(obj)/randstruct.seed: $(gen-randstruct-seed) FORCE
$(call if_changed,create_randstruct_seed)
is equivalent to:
$(obj)/randstruct_hash.h: $(gen-randstruct-seed) FORCE
$(call if_changed,create_randstruct_seed)
$(obj)/randstruct.seed: $(gen-randstruct-seed) FORCE
$(call if_changed,create_randstruct_seed)
So, this rule is executed twice; for randstruct_hash.h and for randstruct.seed
randstruct_hash.h and randstruct.seed will contain different hash values.
I recommend keeping the current code.
--
Best Regards
Masahiro Yamada
Powered by blists - more mailing lists